From 8d28fd9f9e85f5e0715b3be2337b7843fc4c943f Mon Sep 17 00:00:00 2001 From: YurenHao0426 Date: Wed, 22 Jul 2026 16:54:31 -0500 Subject: results: close stability-margin training screen --- .../kp_innovation_stability_grid/margin_0p03.json | 32986 +++++++++++++++++++ 1 file changed, 32986 insertions(+) create mode 100644 results/kp_innovation_stability_grid/margin_0p03.json (limited to 'results/kp_innovation_stability_grid/margin_0p03.json') diff --git a/results/kp_innovation_stability_grid/margin_0p03.json b/results/kp_innovation_stability_grid/margin_0p03.json new file mode 100644 index 0000000..4fd80d2 --- /dev/null +++ b/results/kp_innovation_stability_grid/margin_0p03.json @@ -0,0 +1,32986 @@ +{ + "first_any_nonfinite_groups": {}, + "first_any_nonfinite_step": null, + "first_nonfinite_by_group": {}, + "first_nonfinite_by_metric": {}, + "first_nonfinite_metrics": [], + "first_training_failure_groups": {}, + "first_training_failure_step": null, + "initial_parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4011031687259674, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.8331910371780396, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.0, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3956821858882904, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.1162850856781006, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "max_steps": 352, + "predictor_every": 0, + "predictor_mode": "closed_form", + "predictor_warmup": { + "closed_form_fit": { + "max_absolute_residual_soma_slope": 0.03337550908327103, + "max_applied_stability_margin": 0.033375516533851624, + "max_positive_residual_soma_slope": 0.0, + "min_residual_soma_slope": -0.03337550908327103, + "mse": 0.0007272035929971659, + "observations": 64, + "residual_traffic_rms_ratio": 1.5995902476703123, + "stability_margin": 0.03 + }, + "first_mse": 0.0007272035929971659, + "last_mse": 0.0007272035929971659, + "post_warmup_traffic_residual_rms_ratio": 1.5995902476703123, + "steps": 1, + "task_loader_state_restored": true + }, + "protocol": "kp_mixed_traffic_nonfinite_diagnosis_v1", + "provenance": { + "git_commit": "0ae690e7b3144da2c5b3384cd7169e86529fd26f", + "git_tracked_dirty": false + }, + "rule": "innovation", + "scope": "training_only_no_validation_or_test_evaluation", + "split": { + "dataset": "cifar10", + "input_layout": "NCHW", + "input_shape": [ + 3, + 32, + 32 + ], + "loader_seed": 0, + "normalization_mean": [ + 0.49140000343322754, + 0.4821999967098236, + 0.4465000033378601 + ], + "normalization_std": [ + 0.24699999392032623, + 0.2434999942779541, + 0.26159998774528503 + ], + "split_from_training_only": true, + "split_seed": 2027, + "test_examples": 10000, + "train_examples": 45000, + "training_augmentation": "random_crop_32_padding_4_zero_then_horizontal_flip_p0.5", + "validation_class_counts": { + "0": 500, + "1": 500, + "2": 500, + "3": 500, + "4": 500, + "5": 500, + "6": 500, + "7": 500, + "8": 500, + "9": 500 + }, + "validation_examples": 5000, + "validation_index_sha256": "8328b206a97c420e49e54e3eca4abe3274c4756b084355784ea3fb8059e4515b" + }, + "stability_margin": 0.03, + "test_evaluations": 0, + "traffic_calibration": { + "instruction_rms": [ + 0.008471464738249779, + 0.006617916747927666, + 0.003988116513937712, + 0.0043865954503417015, + 0.002713605063036084, + 0.003158062230795622, + 0.002004207344725728, + 0.003757704282179475, + 0.0027085733599960804, + 0.0030200635083019733, + 0.002056022873148322, + 0.002389939734712243, + 0.0016842411132529378, + 0.0034010226372629404, + 0.002549725119024515, + 0.0030542907770723104, + 0.0020918569061905146, + 0.002605808200314641, + 0.0018916188273578882 + ], + "realized_traffic_instruction_rms_ratio": [ + 4.0, + 4.0, + 4.0, + 4.0, + 4.0, + 4.0, + 3.999999523162842, + 4.0, + 4.000000476837158, + 3.999999761581421, + 4.0, + 4.0, + 3.999999761581421, + 4.0, + 4.000000476837158, + 4.000000476837158, + 4.0, + 4.0, + 4.0 + ], + "target_ratio": 4.0, + "traffic_gain": [ + 0.0443243607878685, + 0.035817068070173264, + 0.013778294436633587, + 0.022762222215533257, + 0.007214079145342112, + 0.017218582332134247, + 0.004616298712790012, + 0.02042512409389019, + 0.007327491883188486, + 0.01611751690506935, + 0.004703814629465342, + 0.01256294921040535, + 0.003324519144371152, + 0.01805141195654869, + 0.006063441745936871, + 0.016296451911330223, + 0.0044285994954407215, + 0.013874167576432228, + 0.0036007820162922144 + ], + "unscaled_traffic_rms": [ + 0.7644973993301392, + 0.7390796542167664, + 1.157796859741211, + 0.7708553671836853, + 1.504616141319275, + 0.7336404919624329, + 1.736635684967041, + 0.7358984351158142, + 1.478581428527832, + 0.7495108246803284, + 1.7483876943588257, + 0.7609485983848572, + 2.0264477729797363, + 0.7536302804946899, + 1.6820316314697266, + 0.7496824264526367, + 1.8894071578979492, + 0.7512690424919128, + 2.10134220123291 + ] + }, + "trajectory": [ + { + "batch_loss": 2.9134020805358887, + "innovation_rms": 0.027791247511226837, + "instruction_rms": 0.004215397890235521, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.40285658836364746, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.9875131845474243, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 28.124950408935547, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4058851897716522, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.6322960257530212, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.7759182453155518, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.1675523519515991, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 11.67552375793457, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.870853066444397, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 28.70853042602539, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 28.124971389770508, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3794170618057251, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.632293164730072, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.757441282272339, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 0.017318262576895525, + "step": 1, + "teaching_rms": 0.027791247511226837, + "traffic_rms": 0.01680093259933294 + }, + { + "batch_loss": 2.306366443634033, + "innovation_rms": 0.2742800857960511, + "instruction_rms": 0.2727156771548255, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.36302483081817627, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 159.16864013671875, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 25.28965950012207, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4102132022380829, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.5693559646606445, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.304884433746338, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.376168966293335, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 12.399652481079102, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.727872848510742, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 70.75712585449219, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 25.28969955444336, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.387024462223053, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.56935054063797, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.233582496643066, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 0.2727708320848526, + "step": 2, + "teaching_rms": 0.2742800857960511, + "traffic_rms": 0.005487565935528707 + }, + { + "batch_loss": 2.491521120071411, + "innovation_rms": 0.41671248531981364, + "instruction_rms": 0.41603139938179945, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.903782606124878, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 569.750244140625, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 22.760162353515625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4162951409816742, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4571085274219513, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.5809006690979, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.486708641052246, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 16.103092193603516, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 12.09678840637207, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 63.68915939331055, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 22.76021957397461, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4192211627960205, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4571007490158081, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.462057113647461, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 0.4161360134718316, + "step": 3, + "teaching_rms": 0.41671248531981364, + "traffic_rms": 0.009329334142785225 + }, + { + "batch_loss": 3.1429684162139893, + "innovation_rms": 0.1158643752561682, + "instruction_rms": 0.10390036522525907, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.350889205932617, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 2275.425048828125, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 20.483388900756836, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.42041441798210144, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.613958477973938, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 9.629240036010742, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.5017499923706055, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 35.53070068359375, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 17.809185028076172, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 245.2412109375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 20.483461380004883, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.43143633008003235, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.614013671875, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 9.467609405517578, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 0.1046479262229814, + "step": 4, + "teaching_rms": 0.1158643752561682, + "traffic_rms": 0.012500514106978803 + }, + { + "batch_loss": 3.19339919090271, + "innovation_rms": 58.235019530769684, + "instruction_rms": 58.23502365994121, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.714153289794922, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 3589.557373046875, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 28.619752883911133, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4203285276889801, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4958186149597168, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 11.472648620605469, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 9.470046997070312, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 34.26422882080078, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 36.93264389038086, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 220.7170867919922, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 28.619718551635742, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4264024794101715, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.49588438868522644, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 11.272512435913086, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 58.23502402371302, + "step": 5, + "teaching_rms": 58.235019530769684, + "traffic_rms": 0.006512936499391681 + }, + { + "batch_loss": 3.537811279296875, + "innovation_rms": 73.7926624976436, + "instruction_rms": 73.79245564113647, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.04273796081543, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 49540.1484375, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 123.5781478881836, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.41928818821907043, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.5724515318870544, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 13.079181671142578, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 13.74576473236084, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 42.757179260253906, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 56.73013687133789, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 936.6278076171875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 123.5781021118164, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.40073075890541077, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.5723889470100403, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 12.866990089416504, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 73.79246161674129, + "step": 6, + "teaching_rms": 73.7926624976436, + "traffic_rms": 0.02969872165885092 + }, + { + "batch_loss": 3.240792751312256, + "innovation_rms": 0.48002596093466, + "instruction_rms": 0.4786552261801404, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.991260528564453, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 44619.125, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 111.21099090576172, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.418327271938324, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.7266679406166077, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 23.22978973388672, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 17.593910217285156, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 51.383060455322266, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 141.02664184570312, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 842.9650268554688, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 111.21094512939453, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.40254655480384827, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.7268329858779907, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 23.331466674804688, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 0.4787203509413934, + "step": 7, + "teaching_rms": 0.48002596093466, + "traffic_rms": 0.007898683708078637 + }, + { + "batch_loss": 3.735729694366455, + "innovation_rms": 13.730580066377033, + "instruction_rms": 13.73016479270646, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 14.205096244812012, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 212669.921875, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 769.8477172851562, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.41807821393013, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.9059659242630005, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 77.23104858398438, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 26.373794555664062, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 87.79884338378906, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 216.89349365234375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 758.6685180664062, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 769.8477172851562, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.39895960688591003, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.9059500694274902, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 77.23650360107422, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 13.730184271046792, + "step": 8, + "teaching_rms": 13.730580066377033, + "traffic_rms": 0.023129084732984695 + }, + { + "batch_loss": 2.9522016048431396, + "innovation_rms": 12.792351409203924, + "instruction_rms": 12.790617320539493, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 34.6671028137207, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1237373.125, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2613.992431640625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.46945905685424805, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.5507502555847168, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 261.17010498046875, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 41.50479507446289, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 151.30999755859375, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 285.17364501953125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 682.8016357421875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2613.992431640625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4438917338848114, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.550733208656311, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 261.1007385253906, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 12.790702916834718, + "step": 9, + "teaching_rms": 12.792351409203924, + "traffic_rms": 0.04679590655215282 + }, + { + "batch_loss": 4.459118366241455, + "innovation_rms": 1.248058394659339, + "instruction_rms": 1.238184604239556, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 32.624603271484375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1198244.625, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4072.13720703125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.5493706464767456, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.3090943098068237, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 496.4267883300781, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 55.46990966796875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 139.65115356445312, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 345.19439697265625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1122.7869873046875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4072.13720703125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4700230360031128, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.3090760707855225, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 496.357421875, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 1.238506554355396, + "step": 10, + "teaching_rms": 1.248058394659339, + "traffic_rms": 0.028272771096332937 + }, + { + "batch_loss": 13.07931900024414, + "innovation_rms": 4932.03702753456, + "instruction_rms": 4932.037027051426, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 51.465187072753906, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1078420.125, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3664.882568359375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.6687363982200623, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 9.480172157287598, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 774.8396606445312, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 68.03851318359375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 158.4014434814453, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 399.21307373046875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1010.5082397460938, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3664.882568359375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.7183689475059509, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 9.480079650878906, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 774.873291015625, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 4932.037027061946, + "step": 11, + "teaching_rms": 4932.03702753456, + "traffic_rms": 0.010286709689369223 + }, + { + "batch_loss": 8.458208084106445, + "innovation_rms": 4679.551224859599, + "instruction_rms": 4679.551372504901, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 82.41870880126953, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 970578.125, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3298.31689453125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.4372737407684326, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 11.583579063415527, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 1104.67138671875, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 79.35025787353516, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 237.4093475341797, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 447.8298645019531, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 909.4573974609375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3298.31689453125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.5715689659118652, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 11.58348274230957, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 1104.7049560546875, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 4679.551372524324, + "step": 12, + "teaching_rms": 4679.551224859599, + "traffic_rms": 0.01354655256274984 + }, + { + "batch_loss": 7.552237510681152, + "innovation_rms": 484.4487617224929, + "instruction_rms": 484.44852957174726, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 126.60831451416016, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 230681280.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 29558.529296875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.2051479816436768, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 10.425140380859375, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 2955.213134765625, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 89.53083038330078, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 229.4764862060547, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 491.5849914550781, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1233.6102294921875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 29558.529296875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.339433193206787, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 10.425041198730469, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 2955.350341796875, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 484.44853856403, + "step": 13, + "teaching_rms": 484.4487617224929, + "traffic_rms": 0.09334650881356128 + }, + { + "batch_loss": 12.546594619750977, + "innovation_rms": 26657.321958938963, + "instruction_rms": 26657.321958870012, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 116.10694122314453, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 207613152.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 26602.380859375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.8962128162384033, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 9.374384880065918, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 5615.451171875, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 98.69334411621094, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 284.9434814453125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 530.964599609375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1110.2491455078125, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 26602.380859375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.030487537384033, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 9.374281883239746, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 5615.58837890625, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 26657.321958871395, + "step": 14, + "teaching_rms": 26657.321958938963, + "traffic_rms": 0.008714262095196016 + }, + { + "batch_loss": 7.711289405822754, + "innovation_rms": 99711.09270196679, + "instruction_rms": 99711.09270195599, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 143.34132385253906, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 186851840.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 23941.580078125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.626417398452759, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 8.436667442321777, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 8009.609375, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 108.3781509399414, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 428.03607177734375, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 566.40625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 999.2241821289062, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 23941.580078125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.7603673934936523, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 8.436561584472656, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 8009.74658203125, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 99711.09270195618, + "step": 15, + "teaching_rms": 99711.09270196679, + "traffic_rms": 0.006344974499454643 + }, + { + "batch_loss": 17.161775588989258, + "innovation_rms": 8.648362676687954, + "instruction_rms": 8.537200206475967, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 347.6746826171875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 450530496.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 150399.1875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.38568115234375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.815722942352295, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 15042.5205078125, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 147.9202423095703, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 395.4209289550781, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 684.8760986328125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 11728.5146484375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 150399.1875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.519620418548584, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.815982341766357, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 15042.4228515625, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 8.542718178030311, + "step": 16, + "teaching_rms": 8.648362676687954, + "traffic_rms": 0.30700418272848795 + }, + { + "batch_loss": 155.881103515625, + "innovation_rms": 3160.10644287658, + "instruction_rms": 3160.1065239143213, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 312.9072265625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 405477440.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 135357.765625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 8.600763320922852, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 97.68192291259766, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 28578.296875, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 272.49755859375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1245.773193359375, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 1740.4423828125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 10555.6630859375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 135357.765625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 8.672464370727539, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 97.6818618774414, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 28578.19921875, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 3160.1065245371965, + "step": 17, + "teaching_rms": 3160.10644287658, + "traffic_rms": 0.06302831096620834 + }, + { + "batch_loss": 34.335453033447266, + "innovation_rms": 3899.576346666944, + "instruction_rms": 3899.57634264128, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 374.2159729003906, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 364929696.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 121819.125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 17.391141891479492, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 87.90377807617188, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 40760.2109375, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 418.1305236816406, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1456.3297119140625, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 2690.4521484375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 9500.0966796875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 121819.125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 17.46283531188965, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 87.90371704101562, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 40760.11328125, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 3899.5763427110805, + "step": 18, + "teaching_rms": 3899.576346666944, + "traffic_rms": 0.023461197951285446 + }, + { + "batch_loss": 21.595521926879883, + "innovation_rms": 474504.6493998512, + "instruction_rms": 474504.64939984836, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 364.6383361816406, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 328436736.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 109633.1328125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 25.30230712890625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 79.11165618896484, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 51723.5234375, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 549.2001953125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1310.6966552734375, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 3545.4609375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 8550.0869140625, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 109633.1328125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 25.373994827270508, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 79.1115951538086, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 51723.42578125, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 474504.64939984825, + "step": 19, + "teaching_rms": 474504.6493998512, + "traffic_rms": 0.006514353146222026 + }, + { + "batch_loss": 16.156450271606445, + "innovation_rms": 2397.2265999838087, + "instruction_rms": 2397.2265271104934, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 480.5095520019531, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 295593056.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 98664.6484375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 32.42210388183594, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 71.19795227050781, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 61589.98828125, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 667.1629028320312, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1179.626953125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 4314.96875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7695.078125, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 98664.6484375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 32.4937858581543, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 71.19789123535156, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 61589.890625, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 2397.226527119289, + "step": 20, + "teaching_rms": 2397.2265999838087, + "traffic_rms": 0.0067449517638815985 + }, + { + "batch_loss": 26.31068229675293, + "innovation_rms": 167740.33238212974, + "instruction_rms": 167740.33238212366, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 577.4423217773438, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 266033744.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 88792.0234375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 38.81120300292969, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 63.8909797668457, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 70469.1875, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 773.329345703125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1061.6641845703125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 5007.52587890625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6925.5703125, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 88792.0234375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 38.882877349853516, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 63.89092254638672, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 70469.09375, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 167740.33238212374, + "step": 21, + "teaching_rms": 167740.33238212974, + "traffic_rms": 0.006438154068927215 + }, + { + "batch_loss": 11.615644454956055, + "innovation_rms": 71531.02826375452, + "instruction_rms": 71531.02577409564, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 622.8534545898438, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 239430368.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 79905.7734375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 44.561004638671875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 57.49800109863281, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 78459.765625, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 868.879150390625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 955.4977416992188, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 5630.8271484375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6233.01318359375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 79905.7734375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 44.63267135620117, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 57.49794006347656, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 78459.671875, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 71531.02577409621, + "step": 22, + "teaching_rms": 71531.02826375452, + "traffic_rms": 0.00914450637567987 + }, + { + "batch_loss": 57.112552642822266, + "innovation_rms": 891441.6235793108, + "instruction_rms": 891441.6235793101, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 845.2896118164062, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 215487328.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 71907.3515625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 49.67181396484375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 51.10808181762695, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 85650.5, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 954.8739624023438, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 859.9479370117188, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 6191.79833984375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5609.7119140625, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 71907.3515625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 49.743473052978516, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 51.1080207824707, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 85650.40625, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 891441.6235793101, + "step": 23, + "teaching_rms": 891441.6235793108, + "traffic_rms": 0.008005296037976888 + }, + { + "batch_loss": 10.513875007629395, + "innovation_rms": 188469.72392814892, + "instruction_rms": 188469.72392814409, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 1616.259033203125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 193938592.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 64708.046875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 54.27104568481445, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 45.992305755615234, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 92121.3046875, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 1032.269287109375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 773.953125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 6696.67236328125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5048.74072265625, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 64708.046875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 54.34269714355469, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 45.992244720458984, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 92121.2109375, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 188469.7239281443, + "step": 24, + "teaching_rms": 188469.72392814892, + "traffic_rms": 0.008207758892832828 + }, + { + "batch_loss": 51.883277893066406, + "innovation_rms": 131110.51608994542, + "instruction_rms": 131110.5160899247, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2697.659912109375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 174544736.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 58228.03125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 58.40927505493164, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 41.38229751586914, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 97944.109375, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 1101.925048828125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 696.5578002929688, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 7151.05908203125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4543.86669921875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 58228.03125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 58.48092269897461, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 41.38223648071289, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 97944.015625, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 131110.5160899254, + "step": 25, + "teaching_rms": 131110.51608994542, + "traffic_rms": 0.013688997751832032 + }, + { + "batch_loss": 9.0996675491333, + "innovation_rms": 377468.86071374227, + "instruction_rms": 377468.86071374035, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 3621.967529296875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 157090256.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 52395.43359375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 62.13230895996094, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 37.230350494384766, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 103183.65625, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 1164.615234375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 626.9019775390625, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 7560.00732421875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4089.47998046875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 52395.43359375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 62.20395278930664, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 37.23028564453125, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 103183.5625, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 377468.8607137406, + "step": 26, + "teaching_rms": 377468.86071374227, + "traffic_rms": 0.010494789055147346 + }, + { + "batch_loss": 4.8094682693481445, + "innovation_rms": 199763.7817586644, + "instruction_rms": 199763.78175866228, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 3726.2841796875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 141381232.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 47145.5703125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 65.482421875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 33.5010986328125, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 107898.2109375, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 1221.036376953125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 564.2117919921875, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 7928.060546875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3680.531982421875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 47145.57421875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 65.5540542602539, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 33.501033782958984, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 107898.1171875, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 199763.78175866255, + "step": 27, + "teaching_rms": 199763.7817586644, + "traffic_rms": 0.00954790396445418 + }, + { + "batch_loss": 126.79713439941406, + "innovation_rms": 17174.074657231813, + "instruction_rms": 17174.07411807726, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 21018.064453125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 12465562714112.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 424094.46875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 68.49686431884766, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 30.981548309326172, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 112140.234375, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 1291.1641845703125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 701.2784423828125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 8259.30859375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 19507.6953125, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 424094.46875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 68.56848907470703, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 30.981475830078125, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 112140.140625, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 17174.074171363012, + "step": 28, + "teaching_rms": 17174.074657231813, + "traffic_rms": 1.3528792572110375 + }, + { + "batch_loss": 8293.240234375, + "innovation_rms": 522648.45853410865, + "instruction_rms": 522648.4581837878, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 1841935.25, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 394118714884096.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1128437.125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 241.95191955566406, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2447.048095703125, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 115956.9296875, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 11560.9482421875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 112354.6015625, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 2763.4462890625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 83781.0625, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1128437.125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 242.38111877441406, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2447.0478515625, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 115956.8359375, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 522648.4582074745, + "step": 29, + "teaching_rms": 522648.45853410865, + "traffic_rms": 4.976345745394231 + }, + { + "batch_loss": 3555.84521484375, + "innovation_rms": 8.722347938624939, + "instruction_rms": 1.8094406403532988, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 10405205.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 354710544449536.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1015584.875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 462.183837890625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2202.319091796875, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 214405.734375, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 21672.86328125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 101119.140625, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 7659.09326171875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 75402.953125, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1015584.875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 462.6130065917969, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2202.31884765625, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 214405.890625, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 3.3442656479893595, + "step": 30, + "teaching_rms": 8.722347938624939, + "traffic_rms": 2.8128056056991073 + }, + { + "batch_loss": 1037.2821044921875, + "innovation_rms": 2834571.876946507, + "instruction_rms": 2834571.8769464884, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 9364684.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 319239483293696.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 914004.9375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 660.387939453125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1982.0408935546875, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 305806.21875, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 30773.5859375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 91007.2265625, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 14445.359375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 67862.65625, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 914004.9375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 660.8170776367188, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1982.0406494140625, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 305806.375, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 2834571.8769464884, + "step": 31, + "teaching_rms": 2834571.876946507, + "traffic_rms": 0.04670209838729726 + }, + { + "batch_loss": 243.35821533203125, + "innovation_rms": 599383.8765332766, + "instruction_rms": 599383.8765324806, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 8428216.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 287315528253440.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 822573.875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 838.7650146484375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1783.770751953125, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 388063.59375, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 38964.234375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 81906.5, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 20552.998046875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 61076.390625, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 822573.875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 839.1941528320312, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1783.7705078125, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 388063.75, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 599383.8765325262, + "step": 32, + "teaching_rms": 599383.8765332766, + "traffic_rms": 0.23721410284644762 + }, + { + "batch_loss": 10615.115234375, + "innovation_rms": 108215019.80915613, + "instruction_rms": 108215019.80914406, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 7585394.5, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.0661983152399974e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 740277.625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 999.2960205078125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1605.309814453125, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 462091.34375, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 46335.8203125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 73715.8515625, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 26049.873046875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 372321.40625, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 740277.625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 999.72509765625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1605.3095703125, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 462091.5, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 108215019.80914456, + "step": 33, + "teaching_rms": 108215019.80915613, + "traffic_rms": 10.36226978340852 + }, + { + "batch_loss": 80380.6953125, + "innovation_rms": 125778.5302708629, + "instruction_rms": 125778.5258924297, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 69847112.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.988305803458642e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 974258.8125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 1140.424072265625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1411.2799072265625, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 528711.6875, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 59943.7421875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 160281.375, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 44454.01171875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 206770.46875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 974258.8125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 1140.85302734375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1411.2796630859375, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 528711.875, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 125778.52607337348, + "step": 34, + "teaching_rms": 125778.5302708629, + "traffic_rms": 6.748812725873796 + }, + { + "batch_loss": 521945.8125, + "innovation_rms": 52323267396.659195, + "instruction_rms": 52323267396.659195, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 62862400.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.7894752746523853e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4971634.5, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 1267.4278564453125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4198.9521484375, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 588664.75, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 80431.640625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 459717.0, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 28009.716796875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 694392.375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4971634.5, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 1267.8568115234375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4198.9521484375, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 588664.9375, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 52323267396.659195, + "step": 35, + "teaching_rms": 52323267396.659195, + "traffic_rms": 19.66077170355208 + }, + { + "batch_loss": 271087.25, + "innovation_rms": 1709036255762.931, + "instruction_rms": 1709036255762.931, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 56576160.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.6105276956475392e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4469686.5, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 1381.7186279296875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3595.8193359375, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 944118.4375, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 121806.171875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 413745.28125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 87480.5390625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 624953.125, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4469686.5, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 1382.1474609375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3595.8193359375, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 944118.5625, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 1709036255762.931, + "step": 36, + "teaching_rms": 1709036255762.931, + "traffic_rms": 6.7912284770662845 + }, + { + "batch_loss": 9415.4609375, + "innovation_rms": 1644879652.2061746, + "instruction_rms": 1644879652.2061746, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 50918544.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.4494748401834394e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4022623.25, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 1484.5665283203125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3231.38427734375, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 1346380.75, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 159043.25, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 372370.75, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 143726.328125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 562457.8125, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4022623.25, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 1484.9952392578125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3231.38427734375, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 1346380.875, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 1644879652.2061746, + "step": 37, + "teaching_rms": 1644879652.2061746, + "traffic_rms": 1.187229421346629 + }, + { + "batch_loss": 637.241943359375, + "innovation_rms": 1096235934.372859, + "instruction_rms": 1096235934.372859, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 45826688.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.3045273389852262e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3620226.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 1577.11474609375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2908.143798828125, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 1708403.375, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 192556.609375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 335133.65625, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 194347.53125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 506212.03125, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3620226.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 1577.54345703125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2908.143798828125, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 1708403.5, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 1096235934.372859, + "step": 38, + "teaching_rms": 1096235934.372859, + "traffic_rms": 3.9002592756564134 + }, + { + "batch_loss": 7307.40673828125, + "innovation_rms": 281973332.8471675, + "instruction_rms": 281973332.8471674, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 41244020.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.1740745793168998e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3258032.5, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 1660.392333984375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2617.197998046875, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 2034206.625, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 222718.640625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 301620.28125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 239906.609375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 455590.8125, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3258032.5, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 1660.821044921875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2617.197998046875, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 2034206.75, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 281973332.8471674, + "step": 39, + "teaching_rms": 281973332.8471675, + "traffic_rms": 2.044278410582729 + }, + { + "batch_loss": 230.920654296875, + "innovation_rms": 4678648207.37208, + "instruction_rms": 4678648207.37208, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 37119616.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.056667095615406e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2932025.75, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 1809.628662109375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2355.320556640625, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 2327409.25, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 249864.46875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 271458.25, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 280909.78125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 410031.71875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2932025.75, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 1809.53125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2355.320556640625, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 2327409.25, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 4678648207.37208, + "step": 40, + "teaching_rms": 4678648207.37208, + "traffic_rms": 1.2121148223786067 + }, + { + "batch_loss": 31762.796875, + "innovation_rms": 7326292460.262199, + "instruction_rms": 7326292460.262199, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 33407654.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 9.510004204136038e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2638590.25, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 2020.88232421875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2112.53662109375, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 2591268.25, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 274295.71875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 244312.421875, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 317812.625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 369028.53125, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2638590.25, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 2020.784912109375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2112.53662109375, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 2591268.25, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7326292460.262199, + "step": 41, + "teaching_rms": 7326292460.262199, + "traffic_rms": 0.5155506315404988 + }, + { + "batch_loss": 12280.7861328125, + "innovation_rms": 10533069480.455685, + "instruction_rms": 10533069480.455685, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 30066888.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 8.559003955521126e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2374472.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 2210.712890625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1898.306640625, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 2828715.5, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 296283.84375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 219881.171875, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 351025.1875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 332125.65625, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2374472.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 2210.615478515625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1898.306640625, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 2828715.5, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 10533069480.455685, + "step": 42, + "teaching_rms": 10533069480.455685, + "traffic_rms": 0.3814607712067667 + }, + { + "batch_loss": 15.086641311645508, + "innovation_rms": 2050028.91916594, + "instruction_rms": 2050028.9191658907, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 27060200.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.703103731767706e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2136742.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 2381.538330078125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1708.2548828125, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 3042389.75, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 316073.15625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 197893.046875, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 380916.5, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 298913.09375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2136742.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 2381.44091796875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1708.2548828125, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 3042389.75, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 2050028.9191659065, + "step": 43, + "teaching_rms": 2050028.91916594, + "traffic_rms": 0.25354018047735316 + }, + { + "batch_loss": 21681.708984375, + "innovation_rms": 148828236.5599722, + "instruction_rms": 148828236.55997217, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 24354180.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.932793186792243e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1922763.5, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 2535.257568359375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1537.191162109375, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 3234666.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 333883.53125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 178103.734375, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 407818.6875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 269021.78125, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1922763.5, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 2535.16015625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1537.191162109375, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 3234666.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 148828236.55997217, + "step": 44, + "teaching_rms": 148828236.5599722, + "traffic_rms": 0.42293929593634483 + }, + { + "batch_loss": 2436.09716796875, + "innovation_rms": 57932495.98792688, + "instruction_rms": 57932495.98792688, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 21918762.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.239513739264e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1730163.625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 2673.579345703125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1383.218505859375, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 3407682.25, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 349912.875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 160293.359375, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 432030.65625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 242119.59375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1730163.625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 2673.48193359375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1383.218505859375, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 3407682.25, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 57932495.98792688, + "step": 45, + "teaching_rms": 57932495.98792688, + "traffic_rms": 0.23322496978847904 + }, + { + "batch_loss": 12.205883979797363, + "innovation_rms": 185332.17738824905, + "instruction_rms": 185332.1773797239, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 19726886.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.6155623653376e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1556806.5, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 2798.042236328125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1244.6292724609375, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 3563363.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 364339.28125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 144264.015625, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 453821.40625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 217907.625, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1556806.5, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 2797.94482421875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1244.6292724609375, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 3563363.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 185332.17737990708, + "step": 46, + "teaching_rms": 185332.17738824905, + "traffic_rms": 0.26055376809742753 + }, + { + "batch_loss": 1941703.125, + "innovation_rms": 268195904690.025, + "instruction_rms": 268195904690.025, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 17754198.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.05400612880384e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1400769.5, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 2879.638671875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1992.3671875, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 3703440.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 377323.03125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 129837.609375, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 473433.09375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 196116.859375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1400769.5, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 2879.541259765625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1992.367431640625, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 3703440.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 268195904690.025, + "step": 47, + "teaching_rms": 268195904690.025, + "traffic_rms": 0.553397681970421 + }, + { + "batch_loss": 912708.125, + "innovation_rms": 29904884219.747295, + "instruction_rms": 29904884219.747295, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 15978778.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.548605301175091e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1646047.875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 2939.28125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5987.8271484375, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 3829472.25, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 389008.40625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116853.84375, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 491083.625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 176505.171875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1646047.875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 2939.183837890625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5987.8271484375, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 3829472.25, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 29904884219.747295, + "step": 48, + "teaching_rms": 29904884219.747295, + "traffic_rms": 1.9042874416897249 + }, + { + "batch_loss": 1255296.125, + "innovation_rms": 20438229.777672652, + "instruction_rms": 20438229.77766848, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 74181400.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.093744642208563e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1481521.875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 2991.13134765625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 8888.6943359375, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 3942863.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 399525.25, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 105168.453125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 506969.09375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 158854.65625, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1481521.875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 2991.033935546875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 8888.6943359375, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 3942863.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 20438229.7776686, + "step": 49, + "teaching_rms": 20438229.777672652, + "traffic_rms": 2.2072521350169505 + }, + { + "batch_loss": 1903974.625, + "innovation_rms": 19851579.057827305, + "instruction_rms": 19851579.057822537, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 169664656.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.267239229141811e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1333415.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 3037.7666015625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7763.541015625, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4044875.25, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 408990.40625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 94651.6015625, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 521266.0, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 142969.1875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1333415.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 3037.669189453125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7763.541015625, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4044875.25, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 19851579.05782267, + "step": 50, + "teaching_rms": 19851579.057827305, + "traffic_rms": 2.3741489900013244 + }, + { + "batch_loss": 2190463.25, + "innovation_rms": 2387762127.413154, + "instruction_rms": 2387762127.413154, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 267703856.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.832115959404954e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1200052.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 3079.7080078125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6788.5595703125, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4136645.75, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 417509.0625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 85186.4375, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 534133.25, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 128672.265625, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1200052.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 3079.610595703125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6788.5595703125, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4136645.75, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 2387762127.413154, + "step": 51, + "teaching_rms": 2387762127.413154, + "traffic_rms": 1.5625218729893113 + }, + { + "batch_loss": 4965542.0, + "innovation_rms": 2545415669.384107, + "instruction_rms": 2545415669.384107, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 401632576.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.8820543098165658e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1080087.25, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 3117.42431640625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5521.22314453125, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4219198.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 425175.84375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 83840.453125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 545713.75, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 115805.0390625, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1080087.25, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 3117.326904296875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5521.22314453125, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4219198.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 2545415669.384107, + "step": 52, + "teaching_rms": 2545415669.384107, + "traffic_rms": 2.1072715659253016 + }, + { + "batch_loss": 202927.78125, + "innovation_rms": 809458919.2205951, + "instruction_rms": 809458919.220595, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 476482848.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.380121051290665e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 972062.8125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 3165.904296875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4959.1953125, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4293452.5, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 432075.9375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 91619.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 556136.1875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 104224.53125, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 972062.8125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 3165.89697265625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4959.1953125, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4293452.5, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 809458919.220595, + "step": 53, + "teaching_rms": 809458919.2205951, + "traffic_rms": 1.3654625140803247 + }, + { + "batch_loss": 437543.1875, + "innovation_rms": 1181350975.3742394, + "instruction_rms": 1181350975.3742394, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 511883136.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.110760271843164e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 874801.5625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 3583.695068359375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4437.04052734375, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4360239.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 438286.03125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 93253.3046875, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 565516.375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 93802.078125, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 874801.5625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 3583.687744140625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4437.04052734375, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4360239.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 1181350975.3742394, + "step": 54, + "teaching_rms": 1181350975.3742394, + "traffic_rms": 1.1014219835801182 + }, + { + "batch_loss": 4365.48583984375, + "innovation_rms": 2826992.510787667, + "instruction_rms": 2826992.510784653, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 505791648.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.962044764449669e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 787235.5, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 3959.6708984375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3993.06689453125, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4420303.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 443875.125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 83927.96875, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 573958.5625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 84421.8671875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 787235.5, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 3959.66357421875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3993.06689453125, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4420303.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 2826992.5107848444, + "step": 55, + "teaching_rms": 2826992.510787667, + "traffic_rms": 1.04184035953934 + }, + { + "batch_loss": 52128.76953125, + "innovation_rms": 1337658588.4103713, + "instruction_rms": 1337658588.4103713, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 668493440.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.5616297678604861e+18, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 708418.1875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 4298.00927734375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3593.450439453125, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4474316.5, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 448905.3125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 75535.171875, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 581556.5, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 75979.6796875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 708418.1875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 4298.001953125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3593.450439453125, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4474316.5, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 1337658588.4103713, + "step": 56, + "teaching_rms": 1337658588.4103713, + "traffic_rms": 0.6701106232185092 + }, + { + "batch_loss": 153.29733276367188, + "innovation_rms": 61663161.01280958, + "instruction_rms": 61663161.01280958, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 647877568.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.475234817074266e+18, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 637475.5625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 4602.47119140625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3233.759765625, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4522884.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 453432.46875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 67981.65625, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 588394.6875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 68381.7109375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 637475.5625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 4602.4638671875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3233.759765625, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4522884.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 61663161.01280958, + "step": 57, + "teaching_rms": 61663161.01280958, + "traffic_rms": 0.23537244363636223 + }, + { + "batch_loss": 4344.93310546875, + "innovation_rms": 1239857566566.8445, + "instruction_rms": 1239857566566.8445, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 626219648.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.3913473025551892e+18, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 573620.8125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 4876.44091796875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2910.006103515625, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4566549.5, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 457506.90625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 61183.48828125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 594549.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 61543.5390625, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 573620.8125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 4876.43359375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2910.006103515625, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4566549.5, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 1239857566566.8445, + "step": 58, + "teaching_rms": 1239857566566.8445, + "traffic_rms": 0.21007295950872582 + }, + { + "batch_loss": 9596.78125, + "innovation_rms": 26984700321.19167, + "instruction_rms": 26984700321.19167, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 612894400.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.296182097270014e+18, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 516145.78125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 5122.96484375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2618.5986328125, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4605802.5, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 461173.90625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 55065.13671875, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 600088.0, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 55389.18359375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 516145.78125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 5122.95751953125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2618.5986328125, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4605802.5, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 26984700321.19167, + "step": 59, + "teaching_rms": 26984700321.19167, + "traffic_rms": 0.9950806862960249 + }, + { + "batch_loss": 10506.9619140625, + "innovation_rms": 3109262540.259982, + "instruction_rms": 3109262540.259982, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 631158848.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.2475227978659267e+18, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 464413.09375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 5344.78515625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2356.3056640625, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4641084.5, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 464474.1875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 49558.62109375, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 605073.0, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 49850.265625, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 464413.09375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 5344.77783203125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2356.3056640625, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4641084.5, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 3109262540.259982, + "step": 60, + "teaching_rms": 3109262540.259982, + "traffic_rms": 1.542005736222716 + }, + { + "batch_loss": 99652.6953125, + "innovation_rms": 19326434.329335883, + "instruction_rms": 19326434.32933566, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 734490368.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.2308502158591918e+18, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 417849.03125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 5544.3701171875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2120.218505859375, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4672791.5, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 467444.46875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 44602.7578125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 609559.5, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 44865.23828125, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 417849.03125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 5544.36279296875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2120.218505859375, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4672791.5, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 19326434.32933571, + "step": 61, + "teaching_rms": 19326434.329335883, + "traffic_rms": 1.3815224269608173 + }, + { + "batch_loss": 39307.4453125, + "innovation_rms": 423187557105.0172, + "instruction_rms": 423187557105.0172, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 698487424.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.1082091083490918e+18, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 375937.21875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 5723.94091796875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1907.7188720703125, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4701281.5, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 470117.71875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 40306.16015625, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 613597.375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 40378.71484375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 375937.21875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 5723.93359375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1907.7188720703125, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 4701281.5, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 423187557105.0172, + "step": 62, + "teaching_rms": 423187557105.0172, + "traffic_rms": 0.6199632296168288 + }, + { + "batch_loss": 94481.1796875, + "innovation_rms": 692888275090.4203, + "instruction_rms": 692888275090.4203, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 796846400.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.7052104558462894e+18, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 119389248.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 5885.49755859375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1716.4500732421875, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 11938925.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 472523.625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 36275.54296875, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 617231.4375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 36340.84375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 119389248.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 5885.490234375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1716.4500732421875, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 11938925.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 692888275090.4203, + "step": 63, + "teaching_rms": 692888275090.4203, + "traffic_rms": 1.9122632409068685 + }, + { + "batch_loss": 4687.9599609375, + "innovation_rms": 16039647972.605211, + "instruction_rms": 16039647972.605211, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 718932672.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.5346903585904394e+18, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 107449128.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6030.83935546875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1544.291015625, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 22683838.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 474688.9375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 32647.98828125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 620502.125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 32706.7578125, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 107449128.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6030.83203125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1544.291015625, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 22683838.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 16039647972.605211, + "step": 64, + "teaching_rms": 16039647972.605211, + "traffic_rms": 0.27351372615930647 + }, + { + "batch_loss": 13.232213973999023, + "innovation_rms": 2.3971254569973466, + "instruction_rms": 2.396423775497152, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 647039424.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.3812213502191862e+18, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 96701944.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6161.5869140625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1389.3323974609375, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 32354032.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 476637.71875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 29383.189453125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 623445.75, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 29436.08203125, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 96701944.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6161.57958984375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1389.3323974609375, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 32354032.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 2.396423775497152, + "step": 65, + "teaching_rms": 2.3971254569973466, + "traffic_rms": 0.00020613807818989675 + }, + { + "batch_loss": 11.577930450439453, + "innovation_rms": 2.7198909224812495, + "instruction_rms": 2.7192547182358227, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 582335488.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.2430991877094769e+18, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 87028512.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6279.1982421875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1249.8558349609375, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 41056884.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 478391.625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 26444.869140625, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 626095.0, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 26492.47265625, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 87028512.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6279.19091796875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1249.8558349609375, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 41056884.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 2.7192546572424914, + "step": 66, + "teaching_rms": 2.7198909224812495, + "traffic_rms": 0.00017788072686848792 + }, + { + "batch_loss": 7.672735214233398, + "innovation_rms": 2.7935484801303647, + "instruction_rms": 2.793064995172039, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 524101952.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.1187892964263199e+18, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 78321552.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6384.9853515625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1124.3143310546875, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 48889040.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 479970.15625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 23800.380859375, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 628479.3125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 23843.224609375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 78321552.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6384.97802734375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1124.3143310546875, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 48889040.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 2.7930649357905333, + "step": 67, + "teaching_rms": 2.7935484801303647, + "traffic_rms": 0.00013493305201180085 + }, + { + "batch_loss": 5.158134460449219, + "innovation_rms": 2.797273445401349, + "instruction_rms": 2.7968937429306924, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 471691744.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.0069103461678449e+18, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 70484504.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6480.1298828125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1011.3157958984375, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 55937492.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 481390.8125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 21420.341796875, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 630625.1875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 21458.90234375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 70484504.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6480.12255859375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1011.3157958984375, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 55937492.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 2.7968937429306924, + "step": 68, + "teaching_rms": 2.797273445401349, + "traffic_rms": 9.7826113813053e-05 + }, + { + "batch_loss": 3.546567916870117, + "innovation_rms": 3.2626760591682067, + "instruction_rms": 3.262585249390722, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 424522560.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 9.062193390388511e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 63430460.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6565.6953125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 909.6069946289062, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 62280540.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 482669.40625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 19278.306640625, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 632556.5, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 19313.01171875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 63430460.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6565.68798828125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 909.6069946289062, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 62280540.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 3.2625852239727804, + "step": 69, + "teaching_rms": 3.2626760591682067, + "traffic_rms": 8.492383034989884e-05 + }, + { + "batch_loss": 3.5565946102142334, + "innovation_rms": 1.503123520278364, + "instruction_rms": 1.5029966337501435, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 382070304.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 8.15597384519123e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 57081184.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6642.638671875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 818.0599975585938, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 67988656.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 483820.15625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 17350.474609375, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 634294.6875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 17381.7109375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 57081184.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6642.63134765625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 818.0599975585938, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 67988656.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 1.5029966337501435, + "step": 70, + "teaching_rms": 1.503123520278364, + "traffic_rms": 8.904332346617729e-05 + }, + { + "batch_loss": 3.8751304149627686, + "innovation_rms": 4.818269196306638, + "instruction_rms": 4.818209000872502, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 343863264.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.340376735550013e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 51366264.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6711.8212890625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 735.6594848632812, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 73125280.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 484855.8125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 15615.4267578125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 635859.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 15643.5390625, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 51366264.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6711.81396484375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 735.6594848632812, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 73125280.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 4.818209000872502, + "step": 71, + "teaching_rms": 4.818269196306638, + "traffic_rms": 9.884947512599928e-05 + }, + { + "batch_loss": 4.165862560272217, + "innovation_rms": 5.08922027637132, + "instruction_rms": 5.089129896078254, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 309476928.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.606339336872919e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 46222324.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6774.0185546875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 661.4916381835938, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 77747512.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 485787.90625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 14053.8837890625, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 637267.0, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 14079.1845703125, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 46222324.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6774.01123046875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 661.4916381835938, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 77747512.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 5.089129896078254, + "step": 72, + "teaching_rms": 5.08922027637132, + "traffic_rms": 0.00011512788364478431 + }, + { + "batch_loss": 4.397496700286865, + "innovation_rms": 4.794923112798185, + "instruction_rms": 4.794807121909659, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 278529248.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.94570533446615e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 41592316.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6829.92822265625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 594.7340087890625, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 81906744.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 486626.8125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 12648.4951171875, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 638534.125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 12671.265625, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 41592316.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6829.9208984375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 594.7340087890625, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 81906744.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 4.794807121909659, + "step": 73, + "teaching_rms": 4.794923112798185, + "traffic_rms": 0.00013768399429907403 + }, + { + "batch_loss": 4.935162544250488, + "innovation_rms": 5.145020565374717, + "instruction_rms": 5.1448832957463155, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 250676320.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.3511348010195354e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 37424892.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6880.1787109375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 534.646240234375, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 85649232.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 487381.8125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 11383.6455078125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 639674.5625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 11404.138671875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 37424892.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6880.17138671875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 534.646240234375, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 85649232.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 5.144883279627738, + "step": 74, + "teaching_rms": 5.145020565374717, + "traffic_rms": 0.00016276549668208488 + }, + { + "batch_loss": 5.190221786499023, + "innovation_rms": 4.625468580170603, + "instruction_rms": 4.625262875298332, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 225608688.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.8160213896370586e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 33673836.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6925.33544921875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 480.5618591308594, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 89016616.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 488061.3125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 10245.2802734375, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 640700.9375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 10263.724609375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 33673836.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6925.328125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 480.5618591308594, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 89016616.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 4.625262857368929, + "step": 75, + "teaching_rms": 4.625468580170603, + "traffic_rms": 0.00018383958000573654 + }, + { + "batch_loss": 4.133368015289307, + "innovation_rms": 4.8530575648461065, + "instruction_rms": 4.852895025083289, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 203047824.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.3344191475941376e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 30297550.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6965.9072265625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 431.8811340332031, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 92046368.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 488672.875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 9220.751953125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 641624.6875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 9237.3515625, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 30297550.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 6965.89990234375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 431.8811340332031, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 92046368.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 4.852895025083289, + "step": 76, + "teaching_rms": 4.8530575648461065, + "traffic_rms": 0.00019617551664209915 + }, + { + "batch_loss": 3.521427631378174, + "innovation_rms": 4.706128230806228, + "instruction_rms": 4.706021923555655, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 182743040.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.900977232834724e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 27258590.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7002.35205078125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 388.06414794921875, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 94772224.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 489223.28125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 8298.6767578125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 642456.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 8313.6162109375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 27258590.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7002.3447265625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 388.06414794921875, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 94772224.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 4.706021923555655, + "step": 77, + "teaching_rms": 4.706128230806228, + "traffic_rms": 0.00020049072800498057 + }, + { + "batch_loss": 3.371114492416382, + "innovation_rms": 4.672610795498748, + "instruction_rms": 4.67248423816426, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 164468736.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.510879440831775e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 24523252.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7035.08251953125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 348.6249694824219, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 97224552.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 489718.625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7468.80908203125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 643204.3125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7482.25439453125, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 24523252.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7035.0751953125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 348.6249694824219, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 97224552.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 4.67248423816426, + "step": 78, + "teaching_rms": 4.672610795498748, + "traffic_rms": 0.00019796900818293644 + }, + { + "batch_loss": 3.109663963317871, + "innovation_rms": 4.725532616192604, + "instruction_rms": 4.725399328658857, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 148021856.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.159791634187551e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 22061204.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7064.46923828125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 313.126220703125, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 99430672.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 490164.4375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6721.92822265625, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 643877.6875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6734.02880859375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 22061204.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7064.4619140625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 313.126220703125, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 99430672.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 4.725399328658857, + "step": 79, + "teaching_rms": 4.725532616192604, + "traffic_rms": 0.00018957258177290174 + }, + { + "batch_loss": 2.9990968704223633, + "innovation_rms": 4.901509493289887, + "instruction_rms": 4.901408614918981, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 133219672.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.8438124364090573e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 19845140.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7090.8466796875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 281.1742248535156, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 101415184.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 490565.6875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6049.7353515625, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 644483.75, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6060.6259765625, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 19845140.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7090.83935546875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 281.1742248535156, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 101415184.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 4.901408614918981, + "step": 80, + "teaching_rms": 4.901509493289887, + "traffic_rms": 0.00017728751118882436 + }, + { + "batch_loss": 2.750286817550659, + "innovation_rms": 4.9985145025468745, + "instruction_rms": 4.998440842672328, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 119897704.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.559431141228544e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 17850484.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7114.515625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 252.41461181640625, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 103200232.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 490926.8125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5444.76171875, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 645029.1875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5454.5634765625, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 17850484.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7114.50830078125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 252.41461181640625, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 103200232.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 4.998440842672328, + "step": 81, + "teaching_rms": 4.9985145025468745, + "traffic_rms": 0.00016400621600418385 + }, + { + "batch_loss": 2.6851541996002197, + "innovation_rms": 4.720001240682983, + "instruction_rms": 4.719937130050532, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 107907936.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.3034880271056896e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 16055115.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7135.74658203125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 226.5284423828125, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 104805744.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 491251.8125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4900.28564453125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 645520.125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4909.10693359375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 16055115.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7135.7392578125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 226.52842712402344, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 104805744.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 4.719937130050532, + "step": 82, + "teaching_rms": 4.720001240682983, + "traffic_rms": 0.00015421784302436739 + }, + { + "batch_loss": 2.706390857696533, + "innovation_rms": 4.866454057057827, + "instruction_rms": 4.866370540172784, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 97117144.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.0731391384957747e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 14439122.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7154.783203125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 203.22862243652344, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 106249656.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 491544.3125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4410.2568359375, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 645961.9375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4418.1962890625, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 14439122.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7154.77587890625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 203.22860717773438, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 106249656.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 4.866370540172784, + "step": 83, + "teaching_rms": 4.866454057057827, + "traffic_rms": 0.00015072496009595622 + }, + { + "batch_loss": 2.666552782058716, + "innovation_rms": 5.162129767890629, + "instruction_rms": 5.161987897815347, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 87405432.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.865825190286459e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 12984584.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7171.84423828125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 182.2567596435547, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 107548112.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 491807.5625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3969.23095703125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 646359.5625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3976.37646484375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 12984584.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7171.8369140625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 182.25674438476562, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 107548112.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 5.161987897815347, + "step": 84, + "teaching_rms": 5.162129767890629, + "traffic_rms": 0.00015333141253582707 + }, + { + "batch_loss": 2.8221521377563477, + "innovation_rms": 5.368359635439227, + "instruction_rms": 5.368266567990865, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 78664888.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.6792427056175514e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 11675370.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7187.12744140625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 163.38026428222656, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 108715648.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 492044.5, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3572.307861328125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 646717.4375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3578.73876953125, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 11675370.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7187.1201171875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 163.38023376464844, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 108715648.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 5.36826658343872, + "step": 85, + "teaching_rms": 5.368359635439227, + "traffic_rms": 0.00015983301917612104 + }, + { + "batch_loss": 2.821532964706421, + "innovation_rms": 5.485180214611453, + "instruction_rms": 5.485133638195781, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 70798400.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.511318417875927e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 10496961.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7200.810546875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 146.3897705078125, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 109765344.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 492257.75, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3215.076904296875, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 647039.5, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3220.86474609375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 10496961.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7200.80322265625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 146.38973999023438, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 109765344.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 5.4851336533145005, + "step": 86, + "teaching_rms": 5.485180214611453, + "traffic_rms": 0.0001678555803921571 + }, + { + "batch_loss": 2.752714157104492, + "innovation_rms": 5.556313132344045, + "instruction_rms": 5.556257972549476, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 63718560.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.3601865932682035e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 9436288.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7213.05322265625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 131.0968780517578, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 110708976.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 492449.65625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2893.569091796875, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 647329.375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2898.778076171875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 9436288.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7213.0458984375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 131.09683227539062, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 110708976.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 5.556257972549476, + "step": 87, + "teaching_rms": 5.556313132344045, + "traffic_rms": 0.00017610965227515157 + }, + { + "batch_loss": 2.748995542526245, + "innovation_rms": 6.087034774543276, + "instruction_rms": 6.0869600772458154, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 57346704.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.2241678995816448e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 8481588.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7223.99951171875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117.33195495605469, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 111557136.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 492622.375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2604.212158203125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 647590.25, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2608.900146484375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 8481588.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7223.9921875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117.3319091796875, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 111557136.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.0869600772458154, + "step": 88, + "teaching_rms": 6.087034774543276, + "traffic_rms": 0.00018318586203247268 + }, + { + "batch_loss": 2.6797497272491455, + "innovation_rms": 6.16325770730061, + "instruction_rms": 6.163264795842611, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 51612032.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.1017511096234803e+17, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7622273.5, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7233.779296875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 104.94234466552734, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 112319360.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 492777.8125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2343.790771484375, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 647825.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2348.010009765625, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7622273.5, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7233.77197265625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 104.94230651855469, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 112319360.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.163264795842611, + "step": 89, + "teaching_rms": 6.16325770730061, + "traffic_rms": 0.00018791090047872813 + }, + { + "batch_loss": 2.7976326942443848, + "innovation_rms": 6.3621287554766885, + "instruction_rms": 6.3621369035269435, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 46450828.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 9.915760072510669e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6848814.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7242.5087890625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 93.7906494140625, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 113004240.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 492917.71875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2109.41162109375, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 648036.375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2113.208984375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6848814.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7242.50146484375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 93.79061126708984, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 113004240.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.3621369035269435, + "step": 90, + "teaching_rms": 6.3621287554766885, + "traffic_rms": 0.0001903813976799624 + }, + { + "batch_loss": 2.6023120880126953, + "innovation_rms": 6.6915586080730405, + "instruction_rms": 6.691599414354479, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 41805744.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 8.924184408856986e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6152632.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7250.29248046875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 83.75318908691406, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 113619504.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 493043.625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1898.470458984375, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 648226.5625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1901.8880615234375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6152632.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7250.28515625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 83.75314331054688, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 113619504.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.691599414354479, + "step": 91, + "teaching_rms": 6.6915586080730405, + "traffic_rms": 0.00019004047501469939 + }, + { + "batch_loss": 2.450298309326172, + "innovation_rms": 6.994762748468965, + "instruction_rms": 6.994752012584559, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 37625168.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 8.031765796172595e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5526006.5, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7257.2255859375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 74.71863555908203, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 114172104.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 493156.9375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1708.6234130859375, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 648397.75, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1711.69921875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5526006.5, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7257.21826171875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 74.71859741210938, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 114172104.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.994752024440333, + "step": 92, + "teaching_rms": 6.994762748468965, + "traffic_rms": 0.0001880583912234487 + }, + { + "batch_loss": 2.5072805881500244, + "innovation_rms": 6.704280970986329, + "instruction_rms": 6.70431064102478, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 33862652.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.228588872957952e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4961988.5, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7263.392578125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 66.5867919921875, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 114668304.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 493258.9375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1537.760986328125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 648551.8125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1540.529296875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4961988.5, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7263.38525390625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 66.58675384521484, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 114668304.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.70431064102478, + "step": 93, + "teaching_rms": 6.704280970986329, + "traffic_rms": 0.00018514130613491272 + }, + { + "batch_loss": 2.421393632888794, + "innovation_rms": 6.8431558703500635, + "instruction_rms": 6.843146243033615, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 30476386.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.505729985662157e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4454322.5, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7268.87060546875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 59.2674674987793, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 115113736.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 493350.71875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1383.98486328125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 648690.4375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1386.476318359375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4454322.5, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7268.86328125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 59.26742935180664, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 115113736.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.843146243033615, + "step": 94, + "teaching_rms": 6.8431558703500635, + "traffic_rms": 0.00018168454083461114 + }, + { + "batch_loss": 2.3856616020202637, + "innovation_rms": 6.536769646642458, + "instruction_rms": 6.536758632540967, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 27428748.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.85515711594496e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3997379.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7273.72802734375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 52.679481506347656, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 115513472.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 493433.34375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1245.5863037109375, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 648815.25, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1247.82861328125, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3997379.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7273.720703125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 52.679447174072266, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 115513472.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.536758632540967, + "step": 95, + "teaching_rms": 6.536769646642458, + "traffic_rms": 0.00017827013504081383 + }, + { + "batch_loss": 2.4668610095977783, + "innovation_rms": 6.680495957511319, + "instruction_rms": 6.6804750013567284, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 24685874.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.269641404350464e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3586089.75, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7278.02685546875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 46.7497673034668, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 115872080.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 493507.6875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1121.027587890625, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 648927.5625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1123.0457763671875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3586089.75, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7278.01953125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 46.749732971191406, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 115872080.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.6804750013567284, + "step": 96, + "teaching_rms": 6.680495957511319, + "traffic_rms": 0.0001754678699377362 + }, + { + "batch_loss": 2.4082882404327393, + "innovation_rms": 6.942897179607749, + "instruction_rms": 6.94288501476497, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 22217286.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.742677263915418e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3215893.5, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7281.8232421875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 41.412559509277344, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116193672.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 493574.59375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1008.9248046875, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649028.625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1010.7411499023438, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3215893.5, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7281.81591796875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 41.41252517700195, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116193672.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.942885026709313, + "step": 97, + "teaching_rms": 6.942897179607749, + "traffic_rms": 0.00017340504889759855 + }, + { + "batch_loss": 2.3376383781433105, + "innovation_rms": 7.500890321570429, + "instruction_rms": 7.500868070158768, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 19995558.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.268409365725184e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2882684.75, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7285.1669921875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 36.6086540222168, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116481944.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 493634.8125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 908.0322875976562, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649119.5625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 909.6669921875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2882684.75, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7285.15966796875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 36.608619689941406, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116481944.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.500868070158768, + "step": 98, + "teaching_rms": 7.500890321570429, + "traffic_rms": 0.00017294201591820953 + }, + { + "batch_loss": 2.3640804290771484, + "innovation_rms": 7.191298602043249, + "instruction_rms": 7.191295620313056, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 17996002.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.841568214404301e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2582768.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7288.103515625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 32.28477478027344, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116740224.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 493689.0, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 817.2290649414062, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649201.4375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 818.7002563476562, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2582768.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7288.09619140625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 32.28474044799805, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116740224.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.191295631844803, + "step": 99, + "teaching_rms": 7.191298602043249, + "traffic_rms": 0.0001743959792387359 + }, + { + "batch_loss": 2.441969156265259, + "innovation_rms": 7.6006923661825745, + "instruction_rms": 7.600699058491957, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 16196402.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.457411307064525e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2312817.25, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7290.67333984375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 28.392961502075195, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116971504.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 493737.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 735.5061645507812, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649275.125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 736.8302001953125, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2312817.25, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7290.666015625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 28.392927169799805, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116971504.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.600699069402558, + "step": 100, + "teaching_rms": 7.6006923661825745, + "traffic_rms": 0.0001773259545533088 + }, + { + "batch_loss": 2.40368390083313, + "innovation_rms": 7.346134682714178, + "instruction_rms": 7.346139891186391, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 14576762.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.1116700904587264e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2069838.375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7292.91357421875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 24.890045166015625, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117178488.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 493781.6875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 661.9555053710938, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649341.4375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 663.1471557617188, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2069838.375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7292.90625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 24.890012741088867, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117178488.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.346139891186391, + "step": 101, + "teaching_rms": 7.346134682714178, + "traffic_rms": 0.0001812375067321766 + }, + { + "batch_loss": 2.3088324069976807, + "innovation_rms": 7.514404986027673, + "instruction_rms": 7.514391556442994, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 13119086.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.800503145837363e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1851136.625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7294.85693359375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 21.737171173095703, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117363600.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 493821.1875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 595.7599487304688, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649401.125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 596.8323974609375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1851136.625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7294.849609375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 21.737138748168945, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117363600.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.514391556442994, + "step": 102, + "teaching_rms": 7.514404986027673, + "traffic_rms": 0.00018529983783960432 + }, + { + "batch_loss": 2.373708963394165, + "innovation_rms": 6.9128208891735135, + "instruction_rms": 6.912761439395106, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 11807177.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.5204528097787904e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1654286.5, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7296.53271484375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 18.899368286132812, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117529032.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 493856.75, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 536.1839599609375, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649454.8125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 537.149169921875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1654286.5, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7296.525390625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 18.899335861206055, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117529032.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.912761439395106, + "step": 103, + "teaching_rms": 6.9128208891735135, + "traffic_rms": 0.00018879767070240428 + }, + { + "batch_loss": 2.321204900741577, + "innovation_rms": 6.83821128938339, + "instruction_rms": 6.838138529163291, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 10626459.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.268407593225421e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1477104.875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7297.96826171875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 16.345157623291016, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117676744.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 493888.75, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 482.5655517578125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649503.125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 483.4342346191406, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1477104.875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7297.9609375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 16.345123291015625, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117676744.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.838138529163291, + "step": 104, + "teaching_rms": 6.83821128938339, + "traffic_rms": 0.00019065794482234138 + }, + { + "batch_loss": 2.306959629058838, + "innovation_rms": 7.05702767264259, + "instruction_rms": 7.056979449062498, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 9563813.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.041566748003533e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1317626.75, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7299.18701171875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 14.04620361328125, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117808504.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 493917.5625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 434.3089904785156, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649546.625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 435.0907897949219, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1317626.75, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7299.1796875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 14.04616928100586, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117808504.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.05697946081373, + "step": 105, + "teaching_rms": 7.05702767264259, + "traffic_rms": 0.00019073670972054752 + }, + { + "batch_loss": 2.3066723346710205, + "innovation_rms": 6.893535710821329, + "instruction_rms": 6.893492043706694, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 8607432.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.8374099873038336e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1174083.125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7300.2109375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 11.977004051208496, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117925912.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 493943.5, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 390.8780822753906, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649585.8125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 391.5816955566406, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1174083.125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7300.20361328125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 11.976970672607422, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117925912.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.893492043706694, + "step": 106, + "teaching_rms": 6.893535710821329, + "traffic_rms": 0.00018946108937008672 + }, + { + "batch_loss": 2.368769884109497, + "innovation_rms": 6.629430440052459, + "instruction_rms": 6.629353349833633, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 7746689.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.6536690315231232e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1044882.1875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7301.0595703125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 10.114605903625488, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118030400.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 493966.84375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 351.7902526855469, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649621.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 352.42352294921875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1044882.1875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7301.05224609375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 10.114571571350098, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118030400.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.629353374852122, + "step": 107, + "teaching_rms": 6.629430440052459, + "traffic_rms": 0.00018741543954835425 + }, + { + "batch_loss": 2.3642804622650146, + "innovation_rms": 6.852547648854232, + "instruction_rms": 6.852467440052597, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 6972020.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.4883021605830656e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 928590.875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7301.75048828125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 8.438345909118652, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118123256.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 493987.84375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 316.6112060546875, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649652.75, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 317.18115234375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 928590.875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7301.7431640625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 8.438311576843262, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118123256.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.852467452154544, + "step": 108, + "teaching_rms": 6.852547648854232, + "traffic_rms": 0.00018466704916437834 + }, + { + "batch_loss": 2.3419060707092285, + "innovation_rms": 6.882066668345699, + "instruction_rms": 6.88201079406638, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 6274818.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.339471987474432e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 823919.4375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7302.29931640625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.957943439483643, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118205648.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494006.75, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 284.9500732421875, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649681.3125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 285.4630432128906, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 823919.4375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7302.2919921875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.958189964294434, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118205648.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.88201079406638, + "step": 109, + "teaching_rms": 6.882066668345699, + "traffic_rms": 0.000181859255250394 + }, + { + "batch_loss": 2.4305567741394043, + "innovation_rms": 6.991262565580623, + "instruction_rms": 6.99119507588205, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 5647336.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.2055247350398976e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 729706.9375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7302.72021484375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.17030143737793, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118278616.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494023.75, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 256.4550476074219, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649707.0, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 256.9167175292969, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 729706.9375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7302.712890625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.170547962188721, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118278616.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.991195087743857, + "step": 110, + "teaching_rms": 6.991262565580623, + "traffic_rms": 0.00017972490214511935 + }, + { + "batch_loss": 2.3168745040893555, + "innovation_rms": 6.153435853351742, + "instruction_rms": 6.153358890634045, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 5082602.5, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.0849723044855808e+16, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 644908.375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7303.02587890625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.461362361907959, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118343104.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494039.0625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 230.80953979492188, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649730.125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 231.22503662109375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 644908.375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7303.0185546875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.46160888671875, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118343104.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.153358890634045, + "step": 111, + "teaching_rms": 6.153435853351742, + "traffic_rms": 0.00017837359625423726 + }, + { + "batch_loss": 2.355905055999756, + "innovation_rms": 6.830606561194431, + "instruction_rms": 6.830547243833888, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 4574342.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 9764750955118592.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 568583.1875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7303.22802734375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.8232622146606445, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118399960.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494052.84375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 207.72857666015625, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649750.9375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 208.10252380371094, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 568583.1875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7303.220703125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.8235087394714355, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118399960.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.830547243833888, + "step": 112, + "teaching_rms": 6.830606561194431, + "traffic_rms": 0.00017805420459192307 + }, + { + "batch_loss": 2.3343265056610107, + "innovation_rms": 6.599697540372068, + "instruction_rms": 6.599659165166549, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 4116907.75, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 8788275644858368.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 499884.84375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7303.3369140625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.2489237785339355, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118449952.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494065.25, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 186.95571899414062, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649769.6875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 187.29226684570312, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 499884.84375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7303.32958984375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.249170780181885, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118449952.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.599659165166549, + "step": 113, + "teaching_rms": 6.599697540372068, + "traffic_rms": 0.0001788067401599778 + }, + { + "batch_loss": 2.304056167602539, + "innovation_rms": 6.45386085379656, + "instruction_rms": 6.45382261425515, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 3705217.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 7909448026685440.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 438051.34375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7303.36181640625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.7319769859313965, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118493760.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494076.40625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 168.26014709472656, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649786.5625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 168.5630340576172, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 438051.34375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7303.3544921875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.7322239875793457, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118493760.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.4538226271046195, + "step": 114, + "teaching_rms": 6.45386085379656, + "traffic_rms": 0.00018055676024407628 + }, + { + "batch_loss": 2.323514461517334, + "innovation_rms": 6.964082438903883, + "instruction_rms": 6.96405710943639, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 3334695.25, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 7118502955581440.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 382396.8125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7303.31103515625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.2666876316070557, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118532000.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494086.4375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 151.4341278076172, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649801.75, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 151.70672607421875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 382396.8125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7303.3037109375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.266934394836426, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118532000.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.96405710943639, + "step": 115, + "teaching_rms": 6.964082438903883, + "traffic_rms": 0.00018291254960408577 + }, + { + "batch_loss": 2.335228204727173, + "innovation_rms": 6.799491662180041, + "instruction_rms": 6.799446500820315, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 3001225.75, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 6406652391587840.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 332303.9375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7303.1923828125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.9924120903015137, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118565232.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494095.46875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 136.29071044921875, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649815.375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 136.5360565185547, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 332303.9375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7303.18505859375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.9923160076141357, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118565232.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.799446500820315, + "step": 116, + "teaching_rms": 6.799491662180041, + "traffic_rms": 0.00018546060607765657 + }, + { + "batch_loss": 2.3172643184661865, + "innovation_rms": 6.4817772551770805, + "instruction_rms": 6.481738760175363, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2701103.25, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 5765987152429056.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 287217.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7303.0126953125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.0366623401641846, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118593952.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494103.59375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 122.66163635253906, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649827.6875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 122.8824462890625, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 287217.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7303.00537109375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.0365662574768066, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118593952.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.481738772969491, + "step": 117, + "teaching_rms": 6.4817772551770805, + "traffic_rms": 0.00018740214655085205 + }, + { + "batch_loss": 2.329347610473633, + "innovation_rms": 6.616700019648508, + "instruction_rms": 6.616623995431451, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2430993.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 5189388329811968.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 246635.90625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7302.77783203125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.0764572620391846, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118618616.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494110.90625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 110.39546966552734, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649838.75, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 110.59420013427734, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 246635.90625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7302.7705078125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.0763609409332275, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118618616.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.616623995431451, + "step": 118, + "teaching_rms": 6.616700019648508, + "traffic_rms": 0.00018837874239663666 + }, + { + "batch_loss": 2.323272943496704, + "innovation_rms": 6.576141437976522, + "instruction_rms": 6.57612195582765, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2187893.75, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 4670449443143680.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 210110.453125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7302.49365234375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.112241744995117, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118639624.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494117.5, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 99.35591888427734, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649848.6875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 99.53477478027344, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 210110.453125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7302.486328125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.11214542388916, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118639624.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.57612195582765, + "step": 119, + "teaching_rms": 6.576141437976522, + "traffic_rms": 0.00018814273045506699 + }, + { + "batch_loss": 2.3347325325012207, + "innovation_rms": 6.717809121372995, + "instruction_rms": 6.717763348868136, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 1969104.375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 4203404498829312.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 177235.4375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7302.16455078125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.289669990539551, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118657344.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494123.4375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 89.42032623291016, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649857.625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 89.58129119873047, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 177235.4375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7302.1572265625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.289661407470703, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118657344.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.717763348868136, + "step": 120, + "teaching_rms": 6.717809121372995, + "traffic_rms": 0.00018716177360587685 + }, + { + "batch_loss": 2.317718029022217, + "innovation_rms": 6.453124052987674, + "instruction_rms": 6.453116805915821, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 1772193.875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 3783064102633472.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 147646.15625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7301.79541015625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.6909193992614746, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118672112.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494128.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 80.4782943725586, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649865.6875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 80.62316131591797, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 147646.15625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7301.7880859375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.690911054611206, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118672112.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.453116805915821, + "step": 121, + "teaching_rms": 6.453124052987674, + "traffic_rms": 0.00018557402918585524 + }, + { + "batch_loss": 2.3198153972625732, + "innovation_rms": 7.030523977118126, + "instruction_rms": 7.030505669518461, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 1594974.5, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 3404757746057216.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 121014.3203125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7301.39013671875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.052006721496582, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118684216.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494133.59375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 72.43046569824219, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649872.9375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 72.56084442138672, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 121014.3203125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7301.3828125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.051998615264893, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118684216.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.030505669518461, + "step": 122, + "teaching_rms": 7.030523977118126, + "traffic_rms": 0.00018391121334532077 + }, + { + "batch_loss": 2.32063364982605, + "innovation_rms": 7.29919445251697, + "instruction_rms": 7.2991978030033255, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 1435477.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 3064281998295040.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 97044.4609375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7300.95263671875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.3769450187683105, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118693920.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494137.90625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 65.18741607666016, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649879.4375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 65.30475616455078, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 97044.4609375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7300.9453125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.376936912536621, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118693920.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.299197814364602, + "step": 123, + "teaching_rms": 7.29919445251697, + "traffic_rms": 0.00018258331859138938 + }, + { + "batch_loss": 2.3266499042510986, + "innovation_rms": 6.9046094494050365, + "instruction_rms": 6.904587939290267, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 1291929.25, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 2757853798465536.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 75470.625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7300.48583984375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.669345855712891, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118701464.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494141.8125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 58.66867446899414, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649885.3125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 58.7742805480957, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 75470.625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7300.478515625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.669337749481201, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118701464.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.904587939290267, + "step": 124, + "teaching_rms": 6.9046094494050365, + "traffic_rms": 0.00018203682280998733 + }, + { + "batch_loss": 2.3482472896575928, + "innovation_rms": 7.205223072185747, + "instruction_rms": 7.205221138361617, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 1162736.375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 2482068311244800.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 56053.4140625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7299.99267578125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.932459831237793, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118707072.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494145.3125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 52.80180740356445, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649890.625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 52.8968505859375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 56053.4140625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7299.9853515625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.9324517250061035, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118707072.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.205221138361617, + "step": 125, + "teaching_rms": 7.205223072185747, + "traffic_rms": 0.0001820925668504731 + }, + { + "batch_loss": 2.299316167831421, + "innovation_rms": 7.06156314067772, + "instruction_rms": 7.061536658562976, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 1046462.75, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 2233861480120320.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 38577.36328125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7299.4755859375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.169212818145752, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118710928.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494148.46875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 47.52162551879883, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649895.375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 47.60716247558594, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 38577.36328125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7299.46826171875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.169205188751221, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118710928.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.061536670306625, + "step": 126, + "teaching_rms": 7.06156314067772, + "traffic_rms": 0.00018286788810060976 + }, + { + "batch_loss": 2.302696943283081, + "innovation_rms": 7.002768598332177, + "instruction_rms": 7.002735723704907, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 941816.5, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 2010475264999424.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 22848.53125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7298.9375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.382238864898682, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118713216.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494151.3125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 42.76946258544922, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649899.6875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 42.8464469909668, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 22848.53125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7298.93017578125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.38223123550415, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118713216.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.002735723704907, + "step": 127, + "teaching_rms": 7.002768598332177, + "traffic_rms": 0.0001839908717402859 + }, + { + "batch_loss": 2.3045222759246826, + "innovation_rms": 7.097911601273733, + "instruction_rms": 7.0978818560826165, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 847634.875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1809427711655936.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 8692.3564453125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7298.3798828125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.57390832901001, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118714088.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494153.875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 38.492515563964844, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649903.5625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 38.56180191040039, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 8692.3564453125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7298.37255859375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.573901176452637, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118714088.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.09788186776613, + "step": 128, + "teaching_rms": 7.097911601273733, + "traffic_rms": 0.00018511647531271758 + }, + { + "batch_loss": 2.3170344829559326, + "innovation_rms": 7.073894651574195, + "instruction_rms": 7.07384286060744, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 762871.375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1628484900225024.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4683.375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7297.80517578125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.746355056762695, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118713680.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494156.1875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 34.64326477050781, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649907.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 34.70561981201172, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4683.375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7297.7978515625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.7463483810424805, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118713680.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.07384286060744, + "step": 129, + "teaching_rms": 7.073894651574195, + "traffic_rms": 0.00018599926804300802 + }, + { + "batch_loss": 2.306735038757324, + "innovation_rms": 7.1146445630258714, + "instruction_rms": 7.114593780284939, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 686584.25, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1465636383358976.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 15514.8271484375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7297.21484375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.901500225067139, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118712128.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494158.25, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 31.178937911987305, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649910.1875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 31.235057830810547, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 15514.8271484375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7297.20751953125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.901493072509766, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118712128.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.114593780284939, + "step": 130, + "teaching_rms": 7.1146445630258714, + "traffic_rms": 0.0001862780951649457 + }, + { + "batch_loss": 2.3144664764404297, + "innovation_rms": 6.419682306130442, + "instruction_rms": 6.419607838868094, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 617925.8125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1319072771866624.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 25834.556640625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7296.61083984375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.04107141494751, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118709544.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494160.125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 28.06104278564453, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649913.0, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 28.11155128479004, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 25834.556640625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7296.603515625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.041064262390137, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118709544.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.419607838868094, + "step": 131, + "teaching_rms": 6.419682306130442, + "traffic_rms": 0.00018599312914720353 + }, + { + "batch_loss": 2.3052940368652344, + "innovation_rms": 7.08241734697993, + "instruction_rms": 7.082374025324468, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 556133.25, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1187165467836416.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 35122.0546875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7295.994140625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.166625022888184, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118706032.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494161.8125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 25.25493812561035, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649915.5, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 25.300395965576172, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 35122.0546875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7295.98681640625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.166618347167969, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118706032.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.082374025324468, + "step": 132, + "teaching_rms": 7.08241734697993, + "traffic_rms": 0.000185015729464906 + }, + { + "batch_loss": 2.2620248794555664, + "innovation_rms": 7.446686604524191, + "instruction_rms": 7.4466545310844126, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 500519.9375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1068448947896320.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 43480.44921875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7295.3662109375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.279561996459961, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118701680.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494163.3125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 22.72944450378418, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649917.75, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 22.770355224609375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 43480.44921875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7295.35888671875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.279554843902588, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118701680.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.4466545310844126, + "step": 133, + "teaching_rms": 7.446686604524191, + "traffic_rms": 0.0001838069631073833 + }, + { + "batch_loss": 2.371011734008789, + "innovation_rms": 7.2102318997984245, + "instruction_rms": 7.210179457540445, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 450467.9375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 961604053106688.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 51002.5703125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7294.72802734375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.3811421394348145, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118696576.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494164.65625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 20.456499099731445, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649919.8125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 20.493318557739258, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 51002.5703125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7294.720703125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.3811354637146, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118696576.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.210179457540445, + "step": 134, + "teaching_rms": 7.2102318997984245, + "traffic_rms": 0.00018304860875225523 + }, + { + "batch_loss": 2.383371114730835, + "innovation_rms": 6.675887495918539, + "instruction_rms": 6.67584024858995, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 405421.15625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 865443627663360.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 57771.96875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7294.08056640625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.472500324249268, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118690800.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494165.875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 18.41084861755371, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649921.6875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 18.443986892700195, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 57771.96875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7294.0732421875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.472494125366211, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118690800.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.6758402610120875, + "step": 135, + "teaching_rms": 6.675887495918539, + "traffic_rms": 0.00018221247441891034 + }, + { + "batch_loss": 2.337975263595581, + "innovation_rms": 7.118761526160655, + "instruction_rms": 7.118720395054209, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 364879.03125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 778899231342592.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 63863.8515625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7293.42529296875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.554657936096191, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118684416.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494166.96875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 16.56976318359375, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649923.375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 16.59958839416504, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 63863.8515625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7293.41796875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.554651737213135, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118684416.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.118720395054209, + "step": 136, + "teaching_rms": 7.118761526160655, + "traffic_rms": 0.00018141183427241163 + }, + { + "batch_loss": 2.301238775253296, + "innovation_rms": 6.972193186541392, + "instruction_rms": 6.97216102311542, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 328391.125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 701009328340992.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 69345.90625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7292.76220703125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.628534317016602, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118677480.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494167.96875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 14.912786483764648, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649924.875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 14.939629554748535, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 69345.90625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7292.7548828125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.628528594970703, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118677480.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.97216102311542, + "step": 137, + "teaching_rms": 6.972193186541392, + "traffic_rms": 0.000181246711164748 + }, + { + "batch_loss": 2.3048219680786133, + "innovation_rms": 7.438790543904323, + "instruction_rms": 7.438740688511563, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 295552.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 630908415639552.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 74279.0625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7292.0927734375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.6949567794799805, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118670056.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494168.84375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 13.421507835388184, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649926.25, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 13.445666313171387, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 74279.0625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7292.08544921875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.694951057434082, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118670056.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.438740699659713, + "step": 138, + "teaching_rms": 7.438790543904323, + "traffic_rms": 0.00018192536499705603 + }, + { + "batch_loss": 2.31916880607605, + "innovation_rms": 6.646106939239022, + "instruction_rms": 6.6460664747967115, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 265996.8125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 567817560653824.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 78718.1640625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7291.41748046875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.754670143127441, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118662184.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494169.65625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 12.079357147216797, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649927.4375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 12.101099014282227, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 78718.1640625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7291.41015625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.754664421081543, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118662184.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.6460664747967115, + "step": 139, + "teaching_rms": 6.646106939239022, + "traffic_rms": 0.00018303443150087612 + }, + { + "batch_loss": 2.3096160888671875, + "innovation_rms": 7.357035523421746, + "instruction_rms": 7.357025715970476, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 239397.125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 511035811299328.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 82712.5625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7290.73681640625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.808344841003418, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118653912.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494170.375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 10.871420860290527, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649928.5, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 10.89098834991455, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 82712.5625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7290.7294921875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.8083391189575195, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118653912.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.357025715970476, + "step": 140, + "teaching_rms": 7.357035523421746, + "traffic_rms": 0.00018417299314360983 + }, + { + "batch_loss": 2.3113739490509033, + "innovation_rms": 7.3710232739502946, + "instruction_rms": 7.3710130239556575, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 215457.40625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 459932243591168.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 86306.6953125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7290.05126953125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.856583595275879, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118645280.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494171.03125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 9.784278869628906, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649929.5, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 9.801889419555664, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 86306.6953125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7290.0439453125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.8565778732299805, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118645280.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.3710130239556575, + "step": 141, + "teaching_rms": 7.3710232739502946, + "traffic_rms": 0.0001855635743541069 + }, + { + "batch_loss": 2.3135807514190674, + "innovation_rms": 6.6724355379330245, + "instruction_rms": 6.672372566187453, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 193911.671875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 413939015876608.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 89540.546875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7289.361328125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.899930000305176, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118636328.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494171.625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 8.805850982666016, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649930.375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 8.821700096130371, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 89540.546875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7289.35400390625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.899924278259277, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118636328.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.672372566187453, + "step": 142, + "teaching_rms": 6.6724355379330245, + "traffic_rms": 0.0001869685455571913 + }, + { + "batch_loss": 2.327869415283203, + "innovation_rms": 7.214495935635864, + "instruction_rms": 7.214508902642965, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 174520.5, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 372545127710720.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 92450.125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7288.66748046875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.938872814178467, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118627080.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494172.15625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.925265789031982, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649931.1875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.939529895782471, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 92450.125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7288.66015625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.938867092132568, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118627080.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.214508902642965, + "step": 143, + "teaching_rms": 7.214495935635864, + "traffic_rms": 0.00018756397214223856 + }, + { + "batch_loss": 2.3173410892486572, + "innovation_rms": 6.604074818469429, + "instruction_rms": 6.604036961720399, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 157068.453125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 335290614939648.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 95067.8203125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7287.97021484375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.973852157592773, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118617576.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494172.625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.132739067077637, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649931.875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.1455769538879395, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 95067.8203125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7287.962890625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.973846435546875, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118617576.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.604036961720399, + "step": 144, + "teaching_rms": 6.604074818469429, + "traffic_rms": 0.00018756098103314345 + }, + { + "batch_loss": 2.319704532623291, + "innovation_rms": 6.470993235599606, + "instruction_rms": 6.470933825062076, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 141361.609375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 301761550090240.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 97422.796875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7287.26953125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.005263805389404, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118607832.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494173.0625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.419465065002441, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649932.5, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.431019306182861, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 97422.796875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7287.26220703125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.005257606506348, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118607832.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.470933825062076, + "step": 145, + "teaching_rms": 6.470993235599606, + "traffic_rms": 0.0001867386459931176 + }, + { + "batch_loss": 2.312774896621704, + "innovation_rms": 6.8193961195707535, + "instruction_rms": 6.81936522997629, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 127225.4453125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 271585395081216.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 99541.296875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7286.56640625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.033463954925537, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118597880.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494173.4375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.777518272399902, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649933.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.787917137145996, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 99541.296875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7286.55908203125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.033458232879639, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118597880.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.81936522997629, + "step": 146, + "teaching_rms": 6.8193961195707535, + "traffic_rms": 0.00018544303487903413 + }, + { + "batch_loss": 2.3247907161712646, + "innovation_rms": 6.861331635145091, + "instruction_rms": 6.861290553998377, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 114502.8984375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 244426857250816.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 101446.953125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7285.8603515625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.058773994445801, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118587736.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494173.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.199766159057617, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649933.5625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.209125518798828, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 101446.953125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7285.85302734375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.058768272399902, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118587736.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.861290553998377, + "step": 147, + "teaching_rms": 6.861331635145091, + "traffic_rms": 0.0001842958719622479 + }, + { + "batch_loss": 2.327329635620117, + "innovation_rms": 7.300601730704977, + "instruction_rms": 7.300575091903722, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 103052.609375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 219984164814848.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 103161.03125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7285.15234375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.081482410430908, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118577416.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494174.09375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.6797895431518555, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649934.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.688212871551514, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 103161.03125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7285.14501953125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.08147668838501, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118577416.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.300575114621989, + "step": 148, + "teaching_rms": 7.300601730704977, + "traffic_rms": 0.0001832040139532615 + }, + { + "batch_loss": 2.3139491081237793, + "innovation_rms": 6.841698894185613, + "instruction_rms": 6.841637526743438, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 92747.3515625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 197985744977920.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 104702.6640625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7284.4423828125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.10184907913208, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118566944.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494174.375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.21181058883667, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649934.5, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.219391345977783, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 104702.6640625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7284.43505859375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.101843357086182, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118566944.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.841637526743438, + "step": 149, + "teaching_rms": 6.841698894185613, + "traffic_rms": 0.00018268739452353494 + }, + { + "batch_loss": 2.2990522384643555, + "innovation_rms": 6.608962557780383, + "instruction_rms": 6.6089537739354, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 83472.6171875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 178187170480128.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 106089.09375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7283.73046875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.120108127593994, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118556336.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494174.625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.7906293869018555, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649934.875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.797452211380005, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 106089.09375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7283.72314453125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.120102405548096, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118556336.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.6089537739354, + "step": 150, + "teaching_rms": 6.608962557780383, + "traffic_rms": 0.0001824804022338585 + }, + { + "batch_loss": 2.303906202316284, + "innovation_rms": 7.157661279151881, + "instruction_rms": 7.1576560299280105, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 75125.3515625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 160368458465280.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 107335.8125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7283.0166015625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.136470317840576, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118545600.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494174.84375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.4115662574768066, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649935.1875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.4177069664001465, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 107335.8125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7283.00927734375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.1364641189575195, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118545600.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.1576560299280105, + "step": 151, + "teaching_rms": 7.157661279151881, + "traffic_rms": 0.00018277427301021046 + }, + { + "batch_loss": 2.33156156539917, + "innovation_rms": 6.385262725809207, + "instruction_rms": 6.385195883488212, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 67612.8125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 144331604230144.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 108456.7890625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7282.30126953125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.151124477386475, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118534752.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494175.0625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.0704095363616943, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649935.5, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.0759360790252686, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 108456.7890625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7282.2939453125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.151118755340576, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118534752.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.385195896475786, + "step": 152, + "teaching_rms": 6.385262725809207, + "traffic_rms": 0.000183553776928683 + }, + { + "batch_loss": 2.3265092372894287, + "innovation_rms": 6.966021440485829, + "instruction_rms": 6.965970251775105, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 60851.53125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 129898442129408.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 109464.5859375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7281.5849609375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.164241790771484, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118523808.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494175.25, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.763368606567383, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649935.75, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.7683424949645996, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 109464.5859375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7281.57763671875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.164236068725586, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118523808.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.965970251775105, + "step": 153, + "teaching_rms": 6.966021440485829, + "traffic_rms": 0.000184033494766314 + }, + { + "batch_loss": 2.316887855529785, + "innovation_rms": 7.02741472405616, + "instruction_rms": 7.027374072127541, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 54766.37890625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 116908598755328.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 110370.5078125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7280.8671875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.175975799560547, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118512768.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494175.40625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.4870316982269287, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649936.0, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.4915082454681396, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 110370.5078125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7280.85986328125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.175970077514648, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118512768.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.027374072127541, + "step": 154, + "teaching_rms": 7.02741472405616, + "traffic_rms": 0.00018446034899199818 + }, + { + "batch_loss": 2.315358877182007, + "innovation_rms": 6.928222924728833, + "instruction_rms": 6.928177620733657, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 49289.7421875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 105217739718656.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 111184.7265625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7280.1484375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.186464309692383, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118501648.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494175.5625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.238328456878662, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649936.25, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.2423572540283203, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 111184.7265625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7280.14111328125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.186459064483643, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118501648.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.928177620733657, + "step": 155, + "teaching_rms": 6.928222924728833, + "traffic_rms": 0.00018498338007124206 + }, + { + "batch_loss": 2.355581045150757, + "innovation_rms": 6.93932036296734, + "instruction_rms": 6.939245087220877, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 44360.76953125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 94695967424512.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 111916.4140625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7279.4287109375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.1958327293396, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118490456.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494175.6875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.014495611190796, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649936.4375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.0181214809417725, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 111916.4140625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7279.42138671875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.195827007293701, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118490456.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.939245087220877, + "step": 156, + "teaching_rms": 6.93932036296734, + "traffic_rms": 0.00018537027359534142 + }, + { + "batch_loss": 2.3116953372955322, + "innovation_rms": 6.416527987443822, + "instruction_rms": 6.41648977200439, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 39924.69140625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 85226369843200.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 112573.8203125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7278.70849609375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.204192161560059, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118479200.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494175.8125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.8130459785461426, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649936.625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.8163093328475952, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 112573.8203125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7278.701171875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.204185962677002, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118479200.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.41648977200439, + "step": 157, + "teaching_rms": 6.416527987443822, + "traffic_rms": 0.00018522083579911668 + }, + { + "batch_loss": 2.2972116470336914, + "innovation_rms": 6.820924501017679, + "instruction_rms": 6.820884477370743, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 35932.22265625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 76703728664576.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 113164.359375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7277.9873046875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.211643695831299, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118467880.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494175.90625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.6317412853240967, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649936.8125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.6346783638000488, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 113164.359375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7277.97998046875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.211637020111084, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118467880.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.820884489528726, + "step": 158, + "teaching_rms": 6.820924501017679, + "traffic_rms": 0.00018474701709287646 + }, + { + "batch_loss": 2.3251304626464844, + "innovation_rms": 6.637552130387107, + "instruction_rms": 6.637491799241756, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 32339.0, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 69033357475840.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 113694.7109375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7277.265625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.218277931213379, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118456512.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.0, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.468567132949829, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649936.9375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.4712104797363281, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 113694.7109375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7277.25830078125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.218271255493164, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118456512.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.637491799241756, + "step": 159, + "teaching_rms": 6.637552130387107, + "traffic_rms": 0.00018421343415213687 + }, + { + "batch_loss": 2.296638250350952, + "innovation_rms": 7.06949436765871, + "instruction_rms": 7.069449992253676, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 29105.099609375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 62130019631104.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 114170.8828125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7276.54296875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.224176406860352, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118445096.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.09375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.3217103481292725, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649937.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.3240894079208374, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 114170.8828125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7276.53564453125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.224169731140137, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118445096.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.069449992253676, + "step": 160, + "teaching_rms": 7.06949436765871, + "traffic_rms": 0.00018355705549608485 + }, + { + "batch_loss": 2.320828437805176, + "innovation_rms": 6.7909260471548025, + "instruction_rms": 6.790898913478453, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 26194.58984375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 55917018087424.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 114598.296875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7275.81982421875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.229413032531738, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118433640.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.1875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.1895393133163452, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649937.1875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.1916804313659668, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 114598.296875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7275.8125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.229405879974365, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118433640.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.790898913478453, + "step": 161, + "teaching_rms": 6.7909260471548025, + "traffic_rms": 0.00018319310008196987 + }, + { + "batch_loss": 2.3603129386901855, + "innovation_rms": 6.508164360507072, + "instruction_rms": 6.50817018726885, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 23575.130859375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 50325314600960.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 114981.828125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7275.09619140625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.234053611755371, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118422144.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.25, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.0705853700637817, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649937.3125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.0725123882293701, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 114981.828125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7275.0888671875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.23404598236084, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118422144.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.50817018726885, + "step": 162, + "teaching_rms": 6.508164360507072, + "traffic_rms": 0.00018320309661951583 + }, + { + "batch_loss": 2.2941160202026367, + "innovation_rms": 7.122924963197873, + "instruction_rms": 7.1228787895062515, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 21217.6171875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 45292783140864.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 115325.8515625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7274.37255859375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.238157749176025, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118410608.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.3125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.9635267853736877, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649937.4375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.9652611017227173, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 115325.8515625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7274.365234375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.238150119781494, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118410608.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.1228787895062515, + "step": 163, + "teaching_rms": 7.122924963197873, + "traffic_rms": 0.00018336950404518272 + }, + { + "batch_loss": 2.339815378189087, + "innovation_rms": 6.389245239321946, + "instruction_rms": 6.389198645406389, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 19095.85546875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 40763505246208.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 115634.328125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7273.6484375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.241779327392578, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118399048.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.8671740889549255, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649937.5, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.8687349557876587, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 115634.328125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7273.64111328125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.241771697998047, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118399048.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.389198645406389, + "step": 164, + "teaching_rms": 6.389245239321946, + "traffic_rms": 0.00018388125022755333 + }, + { + "batch_loss": 2.370469808578491, + "innovation_rms": 7.453937154647178, + "instruction_rms": 7.4539135332647355, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 17186.26953125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 36687153463296.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 115910.796875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7272.923828125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.24496603012085, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118387456.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.4375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.7804566621780396, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649937.5625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.781861424446106, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 115910.796875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7272.91650390625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.244958400726318, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118387456.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.4539135332647355, + "step": 165, + "teaching_rms": 7.453937154647178, + "traffic_rms": 0.00018444940896492632 + }, + { + "batch_loss": 2.2865521907806396, + "innovation_rms": 7.019336749680519, + "instruction_rms": 7.019341273138491, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 15467.642578125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 33018437697536.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116158.453125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7272.19921875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.2477617263793945, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118375840.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.46875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.7024109959602356, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649937.625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.7036752700805664, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116158.453125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7272.19189453125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.247754096984863, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118375840.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.019341273138491, + "step": 166, + "teaching_rms": 7.019336749680519, + "traffic_rms": 0.0001851783058912137 + }, + { + "batch_loss": 2.329319953918457, + "innovation_rms": 7.172059364136785, + "instruction_rms": 7.172017506741768, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 13920.8779296875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 29716593508352.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116380.1875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7271.47412109375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.2502055168151855, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118364200.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.5, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.6321699023246765, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649937.6875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.6333077549934387, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116380.1875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7271.466796875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.250197410583496, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118364200.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.172017506741768, + "step": 167, + "teaching_rms": 7.172059364136785, + "traffic_rms": 0.00018610975346039518 + }, + { + "batch_loss": 2.382507562637329, + "innovation_rms": 6.415824294756265, + "instruction_rms": 6.415796970830924, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 12528.7900390625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 26744933318656.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116578.5859375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7270.7490234375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.2523322105407715, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118352544.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.53125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.5689529180526733, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649937.75, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.5699769854545593, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116578.5859375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7270.74169921875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.252324104309082, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118352544.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.415796970830924, + "step": 168, + "teaching_rms": 6.415824294756265, + "traffic_rms": 0.00018666154384718988 + }, + { + "batch_loss": 2.3402702808380127, + "innovation_rms": 6.47386889313271, + "instruction_rms": 6.473850449014871, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 11275.9111328125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 24070439567360.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116755.984375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7270.0234375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.254173278808594, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118340872.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.5625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.5147121548652649, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649937.8125, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.51297926902771, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116755.984375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7270.01611328125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.2541656494140625, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118340872.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.473850449014871, + "step": 169, + "teaching_rms": 6.47386889313271, + "traffic_rms": 0.00018662634121647344 + }, + { + "batch_loss": 2.325784683227539, + "innovation_rms": 6.880389476983305, + "instruction_rms": 6.88031246237767, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 10148.3203125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 21663395610624.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116914.46875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7269.2978515625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.255757808685303, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118329184.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.59375, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4608518183231354, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649937.875, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4616813361644745, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116914.46875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7269.29052734375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.25575065612793, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118329184.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.880312474430641, + "step": 170, + "teaching_rms": 6.880389476983305, + "traffic_rms": 0.00018600827006168714 + }, + { + "batch_loss": 2.3201255798339844, + "innovation_rms": 6.413705142446999, + "instruction_rms": 6.413654370659298, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 9133.48828125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 19497056468992.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117055.9375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7268.572265625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.257111549377441, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118317480.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4307659864425659, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649937.9375, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.41551318764686584, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117055.9375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7268.56494140625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.257104396820068, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118317480.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.413654383589244, + "step": 171, + "teaching_rms": 6.413705142446999, + "traffic_rms": 0.0001849203750289531 + }, + { + "batch_loss": 2.3079400062561035, + "innovation_rms": 6.415212766605503, + "instruction_rms": 6.415194373858286, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 8220.1396484375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 17547351031808.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117182.09375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7267.8466796875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.2582573890686035, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118305760.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.65625, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.46477848291397095, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.373961865901947, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117182.09375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7267.83935546875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.2582502365112305, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118305760.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.415194386785128, + "step": 172, + "teaching_rms": 6.415212766605503, + "traffic_rms": 0.0001836086364163881 + }, + { + "batch_loss": 2.3207814693450928, + "innovation_rms": 6.993409084683055, + "instruction_rms": 6.993386483910532, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 7398.12548828125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 15792615718912.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117294.4609375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7267.12060546875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.25921630859375, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118294032.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.6875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4287757873535156, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.33656567335128784, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117294.4609375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7267.11328125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.259208679199219, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118294032.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.993386483910532, + "step": 173, + "teaching_rms": 6.993409084683055, + "traffic_rms": 0.00018260002128650985 + }, + { + "batch_loss": 2.312185049057007, + "innovation_rms": 6.939280531366984, + "instruction_rms": 6.939250738141625, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 6658.31298828125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 14213354356736.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117394.421875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7266.39453125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.260006904602051, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118282296.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.71875, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3327052891254425, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.30290910601615906, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117394.421875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7266.38720703125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.259998798370361, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118282296.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.939250738141625, + "step": 174, + "teaching_rms": 6.939280531366984, + "traffic_rms": 0.00018215790675819082 + }, + { + "batch_loss": 2.322450876235962, + "innovation_rms": 6.216901417165734, + "instruction_rms": 6.2168342962658985, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 5992.4814453125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 12792019025920.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117483.203125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7265.66845703125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.260645389556885, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118270544.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.75, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2721283733844757, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2726181745529175, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117483.203125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7265.6611328125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.2606377601623535, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118270544.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.2168342962658985, + "step": 175, + "teaching_rms": 6.216901417165734, + "traffic_rms": 0.00018233919554836513 + }, + { + "batch_loss": 2.337726593017578, + "innovation_rms": 6.845895110326371, + "instruction_rms": 6.845838430766789, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 5393.2333984375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 11512817123328.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117561.9375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7264.9423828125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.261147499084473, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118258784.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.32048001885414124, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.24535635113716125, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117561.9375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7264.93505859375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.261139869689941, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118258784.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.845838430766789, + "step": 176, + "teaching_rms": 6.845895110326371, + "traffic_rms": 0.0001825906279004436 + }, + { + "batch_loss": 2.387932777404785, + "innovation_rms": 6.92842923212916, + "instruction_rms": 6.928388968196657, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 4853.91015625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 10361535201280.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117631.6171875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7264.21630859375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.261527061462402, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118247024.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.47467684745788574, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.22082071006298065, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117631.6171875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7264.208984375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.261518955230713, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118247024.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.928388968196657, + "step": 177, + "teaching_rms": 6.92842923212916, + "traffic_rms": 0.00018309259849949634 + }, + { + "batch_loss": 2.3621747493743896, + "innovation_rms": 6.944213696130634, + "instruction_rms": 6.94417932822675, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 4368.51904296875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 9325381681152.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117693.15625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7263.490234375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.261795997619629, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118235256.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.47550085186958313, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.19873863458633423, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117693.15625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7263.48291015625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.2617878913879395, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118235256.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.94417932822675, + "step": 178, + "teaching_rms": 6.944213696130634, + "traffic_rms": 0.00018370466500286547 + }, + { + "batch_loss": 2.2875139713287354, + "innovation_rms": 6.654865959178554, + "instruction_rms": 6.654822608131091, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 3931.667236328125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 8392843722752.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117747.359375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7262.76416015625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.261965274810791, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118223480.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2655320167541504, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.1788647621870041, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117747.359375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7262.7568359375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.261957168579102, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118223480.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.654822608131091, + "step": 179, + "teaching_rms": 6.654865959178554, + "traffic_rms": 0.00018434620044891308 + }, + { + "batch_loss": 2.3062243461608887, + "innovation_rms": 6.756569842816786, + "instruction_rms": 6.756523535941552, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 3538.50048828125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 7553559298048.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117794.96875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7262.0380859375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.262044906616211, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118211704.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.22237877547740936, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.1609782874584198, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117794.96875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7262.03076171875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.26203727722168, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118211704.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.756523548215349, + "step": 180, + "teaching_rms": 6.756569842816786, + "traffic_rms": 0.00018483035349186068 + }, + { + "batch_loss": 2.315061330795288, + "innovation_rms": 6.804207334811508, + "instruction_rms": 6.804215415798512, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 3184.650390625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 6798203420672.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117836.640625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7261.31201171875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.2620439529418945, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118199920.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.5856512784957886, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14488045871257782, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117836.640625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7261.3046875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.262036323547363, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118199920.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.804215427986281, + "step": 181, + "teaching_rms": 6.804207334811508, + "traffic_rms": 0.000185012249655047 + }, + { + "batch_loss": 2.298367738723755, + "innovation_rms": 7.1584458583171395, + "instruction_rms": 7.158396240957849, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2939.978271484375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 6118383288320.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117872.96875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7260.5859375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.261970520019531, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118188136.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3394421637058258, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.1303924024105072, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117872.96875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7260.57861328125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.261962890625, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118188136.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.158396240957849, + "step": 182, + "teaching_rms": 7.1584458583171395, + "traffic_rms": 0.00018523696044798866 + }, + { + "batch_loss": 2.3443071842193604, + "innovation_rms": 6.0872883202203285, + "instruction_rms": 6.0872109571251904, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2903.08349609375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 5506544959488.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117904.484375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7259.85986328125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.261831760406494, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118176344.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.36622849106788635, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.117353156208992, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117904.484375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7259.8525390625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.261824607849121, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118176344.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.0872109571251904, + "step": 183, + "teaching_rms": 6.0872883202203285, + "traffic_rms": 0.00018543718410668876 + }, + { + "batch_loss": 2.346350908279419, + "innovation_rms": 6.122509062040802, + "instruction_rms": 6.122434378983372, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2828.35546875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 4955890515968.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117931.6640625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7259.1337890625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.261634349822998, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118164552.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.33376917243003845, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.10561783611774445, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117931.6640625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7259.12646484375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.261627197265625, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118164552.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.122434378983372, + "step": 184, + "teaching_rms": 6.122509062040802, + "traffic_rms": 0.00018509929203591665 + }, + { + "batch_loss": 2.330603837966919, + "innovation_rms": 6.968899695219279, + "instruction_rms": 6.968886426406697, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2844.864501953125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 4460301516800.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117954.9453125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7258.40771484375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.2613844871521, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118152760.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3796825110912323, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.09505604952573776, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117954.9453125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7258.400390625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.261376857757568, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118152760.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.968886426406697, + "step": 185, + "teaching_rms": 6.968899695219279, + "traffic_rms": 0.00018429715263629853 + }, + { + "batch_loss": 2.305657386779785, + "innovation_rms": 6.660024655119301, + "instruction_rms": 6.65999353855846, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2918.9794921875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 4014271365120.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117974.71875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7257.681640625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.261086463928223, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118140960.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3711543083190918, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.08555044233798981, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117974.71875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7257.67431640625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.26107931137085, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118140960.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.65999353855846, + "step": 186, + "teaching_rms": 6.660024655119301, + "traffic_rms": 0.0001836904980226516 + }, + { + "batch_loss": 2.3280744552612305, + "innovation_rms": 6.7220673525866115, + "instruction_rms": 6.72204900729865, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2874.620849609375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 3612844228608.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117991.3359375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7256.95556640625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.260745525360107, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118129160.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.33078837394714355, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.07699539512395859, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117991.3359375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7256.9482421875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.260738849639893, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118129160.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.72204900729865, + "step": 187, + "teaching_rms": 6.7220673525866115, + "traffic_rms": 0.00018336954447915896 + }, + { + "batch_loss": 2.3350839614868164, + "innovation_rms": 6.864524098054408, + "instruction_rms": 6.864526912104889, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2810.32177734375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 3251559727104.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118005.1171875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7256.2294921875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.260366439819336, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118117360.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.30717262625694275, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.06929585337638855, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118005.1171875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7256.22216796875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.260359287261963, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118117360.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.864526912104889, + "step": 188, + "teaching_rms": 6.864524098054408, + "traffic_rms": 0.00018335141568567322 + }, + { + "batch_loss": 2.318922519683838, + "innovation_rms": 6.664936532166203, + "instruction_rms": 6.664926865688292, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2805.90234375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 2926403649536.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118016.3359375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7255.50341796875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.259952545166016, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118105560.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.5080845355987549, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.062366265803575516, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118016.3359375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7255.49609375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.259945392608643, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118105560.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.664926865688292, + "step": 189, + "teaching_rms": 6.664936532166203, + "traffic_rms": 0.00018372379474670616 + }, + { + "batch_loss": 2.29748272895813, + "innovation_rms": 6.747587248554058, + "instruction_rms": 6.747542230101712, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2846.233154296875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 2633763389440.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118025.2578125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7254.77734375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.259507179260254, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118093760.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3647289276123047, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.056129638105630875, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118025.2578125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7254.77001953125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.259500503540039, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118093760.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.747542230101712, + "step": 190, + "teaching_rms": 6.747587248554058, + "traffic_rms": 0.00018425500641171234 + }, + { + "batch_loss": 2.3458099365234375, + "innovation_rms": 6.666206322684193, + "instruction_rms": 6.666132506829984, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2890.885498046875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 2370387050496.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118032.1015625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7254.05126953125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.259034156799316, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118081960.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.29276955127716064, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.05051667243242264, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118032.1015625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7254.0439453125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.259027481079102, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118081960.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.666132506829984, + "step": 191, + "teaching_rms": 6.666206322684193, + "traffic_rms": 0.00018492027479174306 + }, + { + "batch_loss": 2.374786615371704, + "innovation_rms": 6.923849856119739, + "instruction_rms": 6.923757501474037, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2841.950439453125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 2133348319232.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118037.0859375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7253.3251953125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.258535861968994, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118070160.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3321911096572876, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.045465003699064255, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118037.0859375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7253.31787109375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.258528709411621, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118070160.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.923757501474037, + "step": 192, + "teaching_rms": 6.923849856119739, + "traffic_rms": 0.00018511429911956546 + }, + { + "batch_loss": 2.337264060974121, + "innovation_rms": 6.473863959953946, + "instruction_rms": 6.473832987222155, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2706.416015625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1920013434880.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118040.390625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7252.599609375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.258014678955078, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118058352.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4984857738018036, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0409185029566288, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118040.390625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7252.59228515625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.258007526397705, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118058352.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.473832987222155, + "step": 193, + "teaching_rms": 6.473863959953946, + "traffic_rms": 0.0001848394258096442 + }, + { + "batch_loss": 2.3261525630950928, + "innovation_rms": 6.585591866592894, + "instruction_rms": 6.585538827373063, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2649.073974609375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1728012091392.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118042.1875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7251.8740234375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.257472991943359, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118046544.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4757539629936218, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.03682665154337883, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118042.1875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7251.86669921875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.257465839385986, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118046544.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.585538827373063, + "step": 194, + "teaching_rms": 6.585591866592894, + "traffic_rms": 0.0001846671763056061 + }, + { + "batch_loss": 2.3077523708343506, + "innovation_rms": 7.095236335106038, + "instruction_rms": 7.095212139843689, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2733.48095703125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1555210829824.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118042.625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7251.1484375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.256912708282471, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118034736.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3041595220565796, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.03314398601651192, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118042.625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7251.14111328125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.256906032562256, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118034736.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.095212139843689, + "step": 195, + "teaching_rms": 7.095236335106038, + "traffic_rms": 0.00018456662709658733 + }, + { + "batch_loss": 2.3637051582336426, + "innovation_rms": 6.590672513371866, + "instruction_rms": 6.590647235889782, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2723.800537109375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1399689707520.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118041.8359375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7250.4228515625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.256336212158203, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118022928.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.32162055373191833, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.029829585924744606, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118041.8359375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7250.41552734375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.256329536437988, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118022928.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.590647235889782, + "step": 196, + "teaching_rms": 6.590672513371866, + "traffic_rms": 0.00018466381037937523 + }, + { + "batch_loss": 2.317683458328247, + "innovation_rms": 6.1233952609520825, + "instruction_rms": 6.123348338678624, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2720.869384765625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1259720671232.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118039.9375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7249.697265625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.255744934082031, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118011128.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.44619643688201904, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0268466267734766, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118039.9375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7249.68994140625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.255737781524658, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 118011128.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.123348352221574, + "step": 197, + "teaching_rms": 6.1233952609520825, + "traffic_rms": 0.00018444035766207952 + }, + { + "batch_loss": 2.2983810901641846, + "innovation_rms": 6.821059076194965, + "instruction_rms": 6.821030846545377, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2665.6337890625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1133748551680.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118037.046875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7248.9716796875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.25514030456543, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117999328.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.5042240619659424, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.024161962792277336, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118037.046875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7248.96435546875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.255132675170898, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117999328.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.821030846545377, + "step": 198, + "teaching_rms": 6.821059076194965, + "traffic_rms": 0.00018391581590349685 + }, + { + "batch_loss": 2.3234493732452393, + "innovation_rms": 6.629796884981649, + "instruction_rms": 6.629771645951061, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2739.925048828125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1020373696512.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118033.265625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7248.24609375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.254523277282715, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117987528.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.48794588446617126, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.021745765581727028, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118033.265625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7248.23876953125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.254515647888184, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117987528.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.629771645951061, + "step": 199, + "teaching_rms": 6.629796884981649, + "traffic_rms": 0.0001836001292213626 + }, + { + "batch_loss": 2.3373961448669434, + "innovation_rms": 7.172954234351531, + "instruction_rms": 7.172917470503258, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2678.84814453125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 918336307200.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118028.6875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7247.5205078125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.253895282745361, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117975728.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.31953567266464233, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.01957118883728981, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118028.6875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7247.51318359375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.25388765335083, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117975728.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.172917470503258, + "step": 200, + "teaching_rms": 7.172954234351531, + "traffic_rms": 0.00018339647152728448 + }, + { + "batch_loss": 2.3462865352630615, + "innovation_rms": 6.205236707250185, + "instruction_rms": 6.205174607000871, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2749.057861328125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 826502676480.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118023.3828125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7246.79541015625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.253257751464844, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117963928.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.26081860065460205, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.01761407032608986, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118023.3828125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7246.7880859375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.2532501220703125, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117963928.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.205174607000871, + "step": 201, + "teaching_rms": 6.205236707250185, + "traffic_rms": 0.00018343155970511332 + }, + { + "batch_loss": 2.3425025939941406, + "innovation_rms": 6.940744129258521, + "instruction_rms": 6.9406862289064355, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2796.465087890625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 743852408832.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118017.4296875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7246.0703125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.25261116027832, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117952128.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.37534722685813904, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.015852663666009903, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118017.4296875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7246.06298828125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.252604007720947, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117952128.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.9406862289064355, + "step": 202, + "teaching_rms": 6.940744129258521, + "traffic_rms": 0.00018328865869596752 + }, + { + "batch_loss": 2.3094093799591064, + "innovation_rms": 6.8640887908500785, + "instruction_rms": 6.864062827794249, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2712.92431640625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 669467148288.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118010.8984375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7245.34521484375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.251956939697266, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117940328.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4552956223487854, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.014267397113144398, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118010.8984375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7245.337890625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.251949310302734, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117940328.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.864062839875753, + "step": 203, + "teaching_rms": 6.8640887908500785, + "traffic_rms": 0.00018322930627307995 + }, + { + "batch_loss": 2.3195858001708984, + "innovation_rms": 6.7295246424482436, + "instruction_rms": 6.729518085896842, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2705.95068359375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 602520420352.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118003.8359375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7244.6201171875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.251295566558838, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117928528.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2502126693725586, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.012840657494962215, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 118003.8359375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7244.61279296875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.251288414001465, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117928528.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.729518085896842, + "step": 204, + "teaching_rms": 6.7295246424482436, + "traffic_rms": 0.00018372919564642947 + }, + { + "batch_loss": 2.306997299194336, + "innovation_rms": 6.311649800283879, + "instruction_rms": 6.311586275529492, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2725.937744140625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 542268391424.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117996.3046875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7243.89501953125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.2506279945373535, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117916728.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11394746601581573, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.011556591838598251, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117996.3046875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7243.8876953125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.250620365142822, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117916728.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.311586275529492, + "step": 205, + "teaching_rms": 6.311649800283879, + "traffic_rms": 0.00018451246767087494 + }, + { + "batch_loss": 2.3084263801574707, + "innovation_rms": 6.443836851210974, + "instruction_rms": 6.4437998771637695, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2762.43212890625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 488041545728.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117988.34375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7243.169921875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.249954700469971, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117904928.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.33907416462898254, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.010400932282209396, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117988.34375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7243.16259765625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.249946594238281, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117904928.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.443799902902682, + "step": 206, + "teaching_rms": 6.443836851210974, + "traffic_rms": 0.0001849143272866901 + }, + { + "batch_loss": 2.317605972290039, + "innovation_rms": 7.059294171103857, + "instruction_rms": 7.0592831510920195, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2739.083740234375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 439237378048.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117980.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7242.44482421875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.249276161193848, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117893128.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.39499014616012573, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.009360838681459427, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117980.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7242.4375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.249268054962158, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117893128.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.0592831510920195, + "step": 207, + "teaching_rms": 7.059294171103857, + "traffic_rms": 0.0001851691767682845 + }, + { + "batch_loss": 2.292736530303955, + "innovation_rms": 6.876109404656992, + "instruction_rms": 6.876062972351626, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2691.07861328125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 395313643520.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117971.3125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7241.7197265625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.248592853546143, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117881328.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4723494052886963, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.00842475425451994, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117971.3125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7241.71240234375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.248584747314453, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117881328.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.876062972351626, + "step": 208, + "teaching_rms": 6.876109404656992, + "traffic_rms": 0.00018555811363456226 + }, + { + "batch_loss": 2.294790506362915, + "innovation_rms": 6.789851850941021, + "instruction_rms": 6.789813464607469, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2742.525390625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 355782262784.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117962.3125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7240.9951171875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.247905731201172, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117869528.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.30257007479667664, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.007582278456538916, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117962.3125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7240.98779296875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.247897148132324, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117869528.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.789813464607469, + "step": 209, + "teaching_rms": 6.789851850941021, + "traffic_rms": 0.00018579250694958336 + }, + { + "batch_loss": 2.312810182571411, + "innovation_rms": 6.599597716415624, + "instruction_rms": 6.599586120103706, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2705.63134765625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 320204046336.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117953.03125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7240.2705078125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.2472147941589355, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117857736.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.18170930445194244, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.006824050564318895, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117953.03125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7240.26318359375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.24720573425293, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117857736.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.599586120103706, + "step": 210, + "teaching_rms": 6.599597716415624, + "traffic_rms": 0.00018575917524488072 + }, + { + "batch_loss": 2.3206348419189453, + "innovation_rms": 6.484438267760016, + "instruction_rms": 6.484368840511182, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2813.971923828125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 288183648256.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117943.5, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7239.5458984375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.246520519256592, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117845944.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3161879777908325, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.00614164536818862, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117943.5, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7239.53857421875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.246510982513428, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117845944.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.484368840511182, + "step": 211, + "teaching_rms": 6.484438267760016, + "traffic_rms": 0.0001853868651747304 + }, + { + "batch_loss": 2.3193728923797607, + "innovation_rms": 6.645564123380986, + "instruction_rms": 6.645503153416238, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2780.591796875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 259365289984.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117933.7421875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7238.8212890625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.245822906494141, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117834152.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4198176860809326, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0055274805054068565, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117933.7421875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7238.81396484375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.245813369750977, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117834152.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.645503153416238, + "step": 212, + "teaching_rms": 6.645564123380986, + "traffic_rms": 0.0001845268595041916 + }, + { + "batch_loss": 2.319516181945801, + "innovation_rms": 6.3498771631435105, + "instruction_rms": 6.349850429778116, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2693.371826171875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 233428762624.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117923.78125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7238.0966796875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.24512243270874, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117822360.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.31893524527549744, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0049747321754693985, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117923.78125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7238.08935546875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.245112895965576, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117822360.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.349850429778116, + "step": 213, + "teaching_rms": 6.3498771631435105, + "traffic_rms": 0.0001836677316974276 + }, + { + "batch_loss": 2.328481674194336, + "innovation_rms": 6.5290215420912965, + "instruction_rms": 6.528990639496344, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2594.984130859375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 210085888000.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117913.6328125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7237.3720703125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.244419574737549, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117810568.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2257899045944214, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.004477258771657944, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117913.6328125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7237.36474609375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.244410037994385, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117810568.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.528990639496344, + "step": 214, + "teaching_rms": 6.5290215420912965, + "traffic_rms": 0.00018320857357342953 + }, + { + "batch_loss": 2.307856798171997, + "innovation_rms": 6.709297173185594, + "instruction_rms": 6.7092704625122375, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2589.169677734375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 189077291008.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117903.3203125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7236.6474609375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.243714809417725, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117798776.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2023727148771286, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.004029532894492149, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117903.3203125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7236.64013671875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.2437052726745605, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117798776.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.7092704625122375, + "step": 215, + "teaching_rms": 6.709297173185594, + "traffic_rms": 0.00018311557804496636 + }, + { + "batch_loss": 2.311102867126465, + "innovation_rms": 6.3056307026533345, + "instruction_rms": 6.305597601550568, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2534.11572265625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 170169565184.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117892.859375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7235.92333984375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.243008136749268, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117786984.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2394673377275467, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0036265794187784195, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117892.859375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7235.916015625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.2429986000061035, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117786984.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.305597601550568, + "step": 216, + "teaching_rms": 6.3056307026533345, + "traffic_rms": 0.00018339741484567666 + }, + { + "batch_loss": 2.3280556201934814, + "innovation_rms": 6.533689442931576, + "instruction_rms": 6.533621793575537, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2561.117919921875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 153152602112.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117882.265625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7235.19921875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.242299556732178, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117775192.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.36320960521698, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0032639214769005775, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117882.265625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7235.19189453125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.242290496826172, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117775192.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.533621793575537, + "step": 217, + "teaching_rms": 6.533689442931576, + "traffic_rms": 0.0001837784286586776 + }, + { + "batch_loss": 2.3137314319610596, + "innovation_rms": 5.931965552932883, + "instruction_rms": 5.931912320535702, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2589.4091796875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 137837346816.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117871.5546875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7234.47509765625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.241589069366455, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117763408.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.39719754457473755, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0029375292360782623, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117871.5546875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7234.4677734375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.241580486297607, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117763408.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 5.931912320535702, + "step": 218, + "teaching_rms": 5.931965552932883, + "traffic_rms": 0.0001840021210951526 + }, + { + "batch_loss": 2.31494140625, + "innovation_rms": 6.841479004512189, + "instruction_rms": 6.841449379857457, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2688.449462890625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 124053610496.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117860.7421875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7233.7509765625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.240877151489258, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117751624.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.35452038049697876, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.002643776126205921, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117860.7421875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7233.74365234375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.240869045257568, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117751624.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.841449379857457, + "step": 219, + "teaching_rms": 6.841479004512189, + "traffic_rms": 0.00018407810171118866 + }, + { + "batch_loss": 2.3405303955078125, + "innovation_rms": 6.293111531133399, + "instruction_rms": 6.293088670281488, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2575.49609375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 111648251904.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117849.828125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7233.02685546875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.240164756774902, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117739840.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.16905219852924347, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.002379398560151458, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117849.828125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7233.01953125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.240156173706055, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117739840.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.293088670281488, + "step": 220, + "teaching_rms": 6.293111531133399, + "traffic_rms": 0.000184384626419789 + }, + { + "batch_loss": 2.278815269470215, + "innovation_rms": 6.281480594126556, + "instruction_rms": 6.281431865751881, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2571.996337890625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 100483424256.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117838.828125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7232.302734375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.239450931549072, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117728056.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.12625938653945923, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0021414586808532476, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117838.828125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7232.29541015625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.239442348480225, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117728056.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.281431865751881, + "step": 221, + "teaching_rms": 6.281480594126556, + "traffic_rms": 0.0001845945242482466 + }, + { + "batch_loss": 2.323031187057495, + "innovation_rms": 6.286699293206527, + "instruction_rms": 6.2866266521289145, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2518.946044921875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 90435084288.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117827.75, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7231.5791015625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.238736152648926, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117716272.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2856203317642212, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0019273128127679229, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117827.75, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7231.57177734375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.238727569580078, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117716272.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.2866266521289145, + "step": 222, + "teaching_rms": 6.286699293206527, + "traffic_rms": 0.000184712714001399 + }, + { + "batch_loss": 2.326785087585449, + "innovation_rms": 6.6410031302170545, + "instruction_rms": 6.640991566936926, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2508.82177734375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 81391575040.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117816.59375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7230.85546875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.238020420074463, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117704488.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.30883872509002686, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0017345815431326628, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117816.59375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7230.84814453125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.238011360168457, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117704488.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.640991566936926, + "step": 223, + "teaching_rms": 6.6410031302170545, + "traffic_rms": 0.00018451288959726983 + }, + { + "batch_loss": 2.306290626525879, + "innovation_rms": 6.705812549749075, + "instruction_rms": 6.705756615394034, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2515.1015625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 73252413440.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117805.375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7230.1318359375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.237303733825684, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117692704.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4544892907142639, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.001561123295687139, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117805.375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7230.12451171875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.237295150756836, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117692704.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.705756615394034, + "step": 224, + "teaching_rms": 6.705812549749075, + "traffic_rms": 0.00018446643171473847 + }, + { + "batch_loss": 2.3016061782836914, + "innovation_rms": 5.88429101474866, + "instruction_rms": 5.884254163354303, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2592.52685546875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 65927172096.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117794.109375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7229.408203125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.236586570739746, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117680928.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.33579695224761963, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0014050109311938286, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117794.109375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7229.40087890625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.236577987670898, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117680928.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 5.884254163354303, + "step": 225, + "teaching_rms": 5.88429101474866, + "traffic_rms": 0.00018460360815582475 + }, + { + "batch_loss": 2.339987277984619, + "innovation_rms": 7.034706263550451, + "instruction_rms": 7.034724036264097, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2577.614501953125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 59334455296.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117782.7890625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7228.6845703125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.235868453979492, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117669152.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.22688287496566772, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.00126450986135751, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117782.7890625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7228.67724609375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.2358598709106445, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117669152.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.034724036264097, + "step": 226, + "teaching_rms": 7.034706263550451, + "traffic_rms": 0.00018459198710061008 + }, + { + "batch_loss": 2.3301796913146973, + "innovation_rms": 6.515890202909416, + "instruction_rms": 6.515851551704165, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2657.809814453125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 53401010176.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117771.421875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7227.9609375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.23514986038208, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117657376.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.34826767444610596, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0011380588402971625, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117771.421875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7227.95361328125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.235141754150391, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117657376.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.515851551704165, + "step": 227, + "teaching_rms": 6.515890202909416, + "traffic_rms": 0.0001850274641569236 + }, + { + "batch_loss": 2.3217670917510986, + "innovation_rms": 6.142840784799293, + "instruction_rms": 6.142800274846185, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2728.700927734375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 48060907520.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117760.0078125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7227.2373046875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.23443078994751, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117645600.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.36414635181427, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0010242529679089785, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117760.0078125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7227.22998046875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.23442268371582, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117645600.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.142800274846185, + "step": 228, + "teaching_rms": 6.142840784799293, + "traffic_rms": 0.00018513920365402994 + }, + { + "batch_loss": 2.3554773330688477, + "innovation_rms": 5.878408366117118, + "instruction_rms": 5.878355749972768, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2730.6611328125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 43254816768.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117748.5703125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7226.51416015625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.233711242675781, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117633824.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.22085221111774445, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0009218276245519519, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117748.5703125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7226.5068359375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.233703136444092, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117633824.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 5.878355749972768, + "step": 229, + "teaching_rms": 5.878408366117118, + "traffic_rms": 0.0001848694007370171 + }, + { + "batch_loss": 2.3077495098114014, + "innovation_rms": 6.61886342987577, + "instruction_rms": 6.618814604481754, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2743.88134765625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 38929334272.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117737.09375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7225.791015625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.2329912185668945, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117622048.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2509346008300781, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0008296448504552245, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117737.09375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7225.78369140625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.232983112335205, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117622048.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.618814604481754, + "step": 230, + "teaching_rms": 6.61886342987577, + "traffic_rms": 0.0001842581517850277 + }, + { + "batch_loss": 2.322310447692871, + "innovation_rms": 6.533493051778612, + "instruction_rms": 6.533432609219039, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2755.914794921875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 35036401664.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117725.5859375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7225.06787109375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.23227071762085, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117610272.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3105246424674988, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0007466803654097021, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117725.5859375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7225.060546875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.232263088226318, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117610272.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.533432609219039, + "step": 231, + "teaching_rms": 6.533493051778612, + "traffic_rms": 0.00018369307453006074 + }, + { + "batch_loss": 2.309081792831421, + "innovation_rms": 6.4399178697758, + "instruction_rms": 6.439867573294263, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2827.60986328125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 31532761088.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117714.046875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7224.3447265625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.231550216674805, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117598504.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2646254301071167, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.000672012334689498, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117714.046875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7224.33740234375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.231542587280273, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117598504.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.439867573294263, + "step": 232, + "teaching_rms": 6.4399178697758, + "traffic_rms": 0.00018329418028559694 + }, + { + "batch_loss": 2.318525552749634, + "innovation_rms": 6.760904144963893, + "instruction_rms": 6.760908399873411, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2876.6806640625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 28379484160.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117702.4921875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7223.62158203125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.23082971572876, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117586736.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.15500348806381226, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0006048110662959516, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117702.4921875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7223.6142578125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.23082160949707, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117586736.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.760908399873411, + "step": 233, + "teaching_rms": 6.760904144963893, + "traffic_rms": 0.00018303098828838246 + }, + { + "batch_loss": 2.3168275356292725, + "innovation_rms": 6.7292046791482525, + "instruction_rms": 6.729161521959976, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2842.26025390625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 25541535744.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117690.9140625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7222.8984375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.230108737945557, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117574968.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.256197452545166, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.000544329930562526, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117690.9140625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7222.89111328125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.230100631713867, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117574968.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.729161534283681, + "step": 234, + "teaching_rms": 6.7292046791482525, + "traffic_rms": 0.00018328362237576053 + }, + { + "batch_loss": 2.3287227153778076, + "innovation_rms": 6.247067250175316, + "instruction_rms": 6.2470350075985746, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2744.501220703125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 22987382784.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117679.3125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7222.17529296875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.229387283325195, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117563200.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2997933626174927, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0004898969200439751, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117679.3125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7222.16796875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.229379653930664, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117563200.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.2470350075985746, + "step": 235, + "teaching_rms": 6.247067250175316, + "traffic_rms": 0.0001837492441732997 + }, + { + "batch_loss": 2.308645009994507, + "innovation_rms": 6.196441997134758, + "instruction_rms": 6.196384479043028, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2735.51708984375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 20688644096.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117667.6953125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7221.45263671875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.228665828704834, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117551432.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.20902198553085327, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0004409072280395776, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117667.6953125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7221.4453125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.228658199310303, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117551432.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.196384492426348, + "step": 236, + "teaching_rms": 6.196441997134758, + "traffic_rms": 0.0001842295262379409 + }, + { + "batch_loss": 2.3069112300872803, + "innovation_rms": 6.1376472824732184, + "instruction_rms": 6.137566392695007, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2755.68408203125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 18619779072.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117656.0625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7220.72998046875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.227944374084473, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117539664.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.1330595761537552, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0003968164965044707, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117656.0625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7220.72265625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.227936744689941, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117539664.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.137566392695007, + "step": 237, + "teaching_rms": 6.1376472824732184, + "traffic_rms": 0.00018448436386109545 + }, + { + "batch_loss": 2.2766213417053223, + "innovation_rms": 6.900341177080988, + "instruction_rms": 6.900326441570825, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2784.502197265625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 16757800960.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117644.421875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7220.00732421875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.227222919464111, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117527896.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.09916031360626221, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0003571348497644067, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117644.421875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7220.0, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.227214813232422, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117527896.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.9003264535888365, + "step": 238, + "teaching_rms": 6.900341177080988, + "traffic_rms": 0.0001843313386504402 + }, + { + "batch_loss": 2.3357088565826416, + "innovation_rms": 6.688564649820762, + "instruction_rms": 6.688502595562416, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2795.29345703125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 15082020864.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117632.765625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7219.28466796875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.226500988006592, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117516136.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.20206661522388458, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.000321421364787966, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117632.765625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7219.27734375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.2264933586120605, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117516136.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.688502595562416, + "step": 239, + "teaching_rms": 6.688564649820762, + "traffic_rms": 0.00018447690183488502 + }, + { + "batch_loss": 2.307896614074707, + "innovation_rms": 6.858145608794735, + "instruction_rms": 6.858115994889053, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2845.046875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 13573818368.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117621.09375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7218.56201171875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.2257795333862305, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117504376.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3033469319343567, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.00028927921084687114, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117621.09375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7218.5546875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.225771427154541, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117504376.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.858115994889053, + "step": 240, + "teaching_rms": 6.858145608794735, + "traffic_rms": 0.00018446783848328954 + }, + { + "batch_loss": 2.2994472980499268, + "innovation_rms": 6.430860813831955, + "instruction_rms": 6.430800169363558, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2784.502197265625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 12216436736.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117609.421875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7217.83935546875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.225057601928711, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117492616.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.26637864112854004, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.00026035128394141793, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117609.421875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7217.83203125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.22504997253418, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117492616.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.430800169363558, + "step": 241, + "teaching_rms": 6.430860813831955, + "traffic_rms": 0.00018461048270987778 + }, + { + "batch_loss": 2.3057610988616943, + "innovation_rms": 6.748828367830764, + "instruction_rms": 6.748811778908734, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2814.651123046875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 10994793472.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117597.734375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7217.11669921875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.224335670471191, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117480856.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2293611615896225, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.000234316146816127, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117597.734375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7217.109375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.22432804107666, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117480856.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.748811778908734, + "step": 242, + "teaching_rms": 6.748828367830764, + "traffic_rms": 0.0001844377714939653 + }, + { + "batch_loss": 2.312324285507202, + "innovation_rms": 6.647552622624515, + "instruction_rms": 6.647536768610025, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2890.61474609375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 9895314432.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117586.046875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7216.39453125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.223613739013672, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117469096.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.26615583896636963, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.00021088452194817364, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117586.046875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7216.38720703125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.223606109619141, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117469096.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.647536768610025, + "step": 243, + "teaching_rms": 6.647552622624515, + "traffic_rms": 0.00018440364541831046 + }, + { + "batch_loss": 2.320770740509033, + "innovation_rms": 6.579197918479375, + "instruction_rms": 6.579150375365365, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2843.51318359375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 8905783296.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117574.34375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7215.67236328125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.222891807556152, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117457336.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.343375563621521, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.00018979606102220714, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117574.34375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7215.6650390625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.222884178161621, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117457336.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.579150375365365, + "step": 244, + "teaching_rms": 6.579197918479375, + "traffic_rms": 0.00018434520165562962 + }, + { + "batch_loss": 2.3078579902648926, + "innovation_rms": 6.726584239544707, + "instruction_rms": 6.726525125357202, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2914.117431640625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 8015204864.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117562.640625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7214.9501953125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.222169399261475, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117445576.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.31434595584869385, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.00017081644909922034, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117562.640625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7214.94287109375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.222162246704102, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117445576.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.726525125357202, + "step": 245, + "teaching_rms": 6.726584239544707, + "traffic_rms": 0.00018427433384850077 + }, + { + "batch_loss": 2.3164799213409424, + "innovation_rms": 6.494619015353291, + "instruction_rms": 6.494610510744442, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2858.816650390625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 7213684224.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117550.9296875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7214.22802734375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.221447467803955, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117433824.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2424509972333908, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.00015373479982372373, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117550.9296875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7214.220703125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.221439838409424, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117433824.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.49461049797567, + "step": 246, + "teaching_rms": 6.494619015353291, + "traffic_rms": 0.00018425241093530226 + }, + { + "batch_loss": 2.315408229827881, + "innovation_rms": 6.245787143233792, + "instruction_rms": 6.2457506574889115, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2767.302001953125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 6492315648.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117539.21875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7213.505859375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.220725059509277, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117422072.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.1437595635652542, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.00013836131256539375, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117539.21875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7213.49853515625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.220717906951904, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117422072.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.2457506574889115, + "step": 247, + "teaching_rms": 6.245787143233792, + "traffic_rms": 0.00018447210561560286 + }, + { + "batch_loss": 2.314312219619751, + "innovation_rms": 6.2621369366422615, + "instruction_rms": 6.262090996399853, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2761.845458984375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 5843084288.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117527.5, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7212.78369140625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.220003128051758, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117410320.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.15292944014072418, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.0001245251769432798, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117527.5, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7212.7763671875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.219995498657227, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117410320.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.262090996399853, + "step": 248, + "teaching_rms": 6.2621369366422615, + "traffic_rms": 0.00018466419180948566 + }, + { + "batch_loss": 2.3249454498291016, + "innovation_rms": 6.479541046079885, + "instruction_rms": 6.479507117555888, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2817.298828125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 5258776064.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117515.78125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7212.0615234375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.21928071975708, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117398568.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.296479195356369, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.00011207265924895182, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117515.78125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7212.05419921875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.219273567199707, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117398568.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.479507117555888, + "step": 249, + "teaching_rms": 6.479541046079885, + "traffic_rms": 0.0001848740128612717 + }, + { + "batch_loss": 2.3020200729370117, + "innovation_rms": 6.62188901029244, + "instruction_rms": 6.6218472324018345, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2808.2890625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 4732898304.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117504.0625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7211.33984375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.2185587882995605, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117386816.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3738989233970642, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.00010086539259646088, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117504.0625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7211.33251953125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.218551158905029, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117386816.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.6218472324018345, + "step": 250, + "teaching_rms": 6.62188901029244, + "traffic_rms": 0.00018488797553311575 + }, + { + "batch_loss": 2.346102476119995, + "innovation_rms": 7.182598156479782, + "instruction_rms": 7.182575573886158, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2859.23095703125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 4259608576.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117492.3359375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7210.6181640625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.217836380004883, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117375064.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.22309842705726624, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 9.077884897124022e-05, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117492.3359375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7210.61083984375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.21782922744751, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117375064.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.1825755854319056, + "step": 251, + "teaching_rms": 7.182598156479782, + "traffic_rms": 0.00018486412004766798 + }, + { + "batch_loss": 2.3220279216766357, + "innovation_rms": 6.770958473952841, + "instruction_rms": 6.770931910049025, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2894.755859375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 3833647616.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117480.609375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7209.896484375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.217114448547363, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117363312.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.214574933052063, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 8.17009640741162e-05, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117480.609375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7209.88916015625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.21710729598999, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117363312.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.770931922296703, + "step": 252, + "teaching_rms": 6.770958473952841, + "traffic_rms": 0.00018479214057439173 + }, + { + "batch_loss": 2.3091213703155518, + "innovation_rms": 6.620617966462752, + "instruction_rms": 6.620573652013762, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2839.15966796875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 3450282752.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117468.875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7209.1748046875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.216392517089844, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117351568.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2440706193447113, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.353086402872577e-05, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117468.875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7209.16748046875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.216385364532471, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117351568.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.620573652013762, + "step": 253, + "teaching_rms": 6.620617966462752, + "traffic_rms": 0.00018469041491353522 + }, + { + "batch_loss": 2.320317506790161, + "innovation_rms": 6.6207642277457985, + "instruction_rms": 6.620715443429827, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2811.588134765625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 3105254400.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117457.140625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7208.453125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.215670585632324, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117339824.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4309018850326538, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.617777398787439e-05, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117457.140625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7208.44580078125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.215663433074951, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117339824.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.620715443429827, + "step": 254, + "teaching_rms": 6.6207642277457985, + "traffic_rms": 0.00018448775985387486 + }, + { + "batch_loss": 2.3193581104278564, + "innovation_rms": 6.680657129230949, + "instruction_rms": 6.680613573105198, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2713.42333984375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 2794728960.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117445.40625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7207.7314453125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.214948654174805, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117328080.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3438892960548401, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.955999586149119e-05, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117445.40625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7207.72412109375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.214941501617432, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117328080.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.680613573105198, + "step": 255, + "teaching_rms": 6.680657129230949, + "traffic_rms": 0.00018419199374119117 + }, + { + "batch_loss": 2.312452554702759, + "innovation_rms": 7.175004099294974, + "instruction_rms": 7.174979767443306, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2717.962646484375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 2515256064.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117433.6640625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7207.01025390625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.214226722717285, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117316336.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.10968003422021866, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.360399518394843e-05, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117433.6640625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7207.0029296875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.214219570159912, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117316336.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.174979767443306, + "step": 256, + "teaching_rms": 7.175004099294974, + "traffic_rms": 0.00018382745354131555 + }, + { + "batch_loss": 2.344525098800659, + "innovation_rms": 6.498895238996828, + "instruction_rms": 6.498828326363946, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2665.90625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 2263730432.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117421.9296875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7206.2890625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.213504791259766, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117304592.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.22959209978580475, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.82435934827663e-05, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117421.9296875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7206.28173828125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.213497638702393, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117304592.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.498828326363946, + "step": 257, + "teaching_rms": 6.498895238996828, + "traffic_rms": 0.00018384705445672735 + }, + { + "batch_loss": 2.3125298023223877, + "innovation_rms": 6.470311280367587, + "instruction_rms": 6.470278637418909, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2715.404052734375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 2037357440.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117410.1953125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7205.56787109375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.212782859802246, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117292848.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.289000004529953, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.341923340689391e-05, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117410.1953125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7205.560546875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.212776184082031, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117292848.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.470278637418909, + "step": 258, + "teaching_rms": 6.470311280367587, + "traffic_rms": 0.00018352259058360295 + }, + { + "batch_loss": 2.3199024200439453, + "innovation_rms": 5.94101369001869, + "instruction_rms": 5.94095636514713, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2638.995361328125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1833621632.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117398.453125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7204.8466796875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.212060928344727, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117281112.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.24490416049957275, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.907730933860876e-05, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117398.453125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7204.83935546875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.212054252624512, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117281112.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 5.94095636514713, + "step": 259, + "teaching_rms": 5.94101369001869, + "traffic_rms": 0.0001833193253485195 + }, + { + "batch_loss": 2.3164913654327393, + "innovation_rms": 6.773040078105713, + "instruction_rms": 6.773018371461168, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2723.488037109375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1650259456.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117386.7109375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7204.12548828125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.211339473724365, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117269376.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.28705790638923645, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.516957804095e-05, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117386.7109375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7204.1181640625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.211332321166992, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117269376.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.773018371461168, + "step": 260, + "teaching_rms": 6.773040078105713, + "traffic_rms": 0.00018312008588175716 + }, + { + "batch_loss": 2.2993764877319336, + "innovation_rms": 6.535402715909009, + "instruction_rms": 6.53534219099546, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2749.637451171875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1485233536.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117374.9765625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7203.404296875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.210618019104004, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117257640.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.17153245210647583, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.1652620236855e-05, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117374.9765625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7203.39697265625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.210610866546631, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117257640.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.53534219099546, + "step": 261, + "teaching_rms": 6.535402715909009, + "traffic_rms": 0.0001832009786402342 + }, + { + "batch_loss": 2.322821855545044, + "innovation_rms": 5.945805835694903, + "instruction_rms": 5.945746870381875, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2743.277587890625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1336710144.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117363.2421875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7202.68310546875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.209896087646484, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117245904.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.24800074100494385, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.8487356757977977e-05, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117363.2421875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7202.67578125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.2098894119262695, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117245904.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 5.945746870381875, + "step": 262, + "teaching_rms": 5.945805835694903, + "traffic_rms": 0.00018334431194472577 + }, + { + "batch_loss": 2.3150768280029297, + "innovation_rms": 6.245703414780859, + "instruction_rms": 6.24567305000367, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2755.0927734375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1203039104.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117351.5, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7201.96240234375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.209174633026123, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117234168.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.21828246116638184, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.563862108218018e-05, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117351.5, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7201.955078125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.20916748046875, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117234168.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.24567305000367, + "step": 263, + "teaching_rms": 6.245703414780859, + "traffic_rms": 0.0001833844640077857 + }, + { + "batch_loss": 2.2875120639801025, + "innovation_rms": 6.590861223836605, + "instruction_rms": 6.590834473742128, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2721.67529296875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 1082735232.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117339.7578125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7201.24169921875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.208453178405762, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117222432.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.21471631526947021, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.3074759155861102e-05, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117339.7578125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7201.234375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.208446025848389, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117222432.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.590834473742128, + "step": 264, + "teaching_rms": 6.590861223836605, + "traffic_rms": 0.0001835888148488851 + }, + { + "batch_loss": 2.290386915206909, + "innovation_rms": 6.237833525252429, + "instruction_rms": 6.237772677999508, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2761.047119140625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 974461696.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117328.0234375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7200.52099609375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.2077317237854, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117210696.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.21137851476669312, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.076728196698241e-05, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117328.0234375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7200.513671875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.207724571228027, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117210696.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.237772677999508, + "step": 265, + "teaching_rms": 6.237833525252429, + "traffic_rms": 0.00018405843812163418 + }, + { + "batch_loss": 2.321544885635376, + "innovation_rms": 6.091268040730858, + "instruction_rms": 6.091235107198598, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2773.807373046875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 877015552.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117316.2890625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7199.80029296875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.207010746002197, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117198968.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.21705204248428345, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.8690552678890526e-05, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117316.2890625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7199.79296875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.207003116607666, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117198968.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.091235107198598, + "step": 266, + "teaching_rms": 6.091268040730858, + "traffic_rms": 0.00018455262002514997 + }, + { + "batch_loss": 2.3172004222869873, + "innovation_rms": 6.345503258154754, + "instruction_rms": 6.345498499526065, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2737.92333984375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 789313984.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117304.5546875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7199.07958984375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.206289291381836, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117187240.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2862128019332886, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.6821497411001474e-05, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117304.5546875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7199.072265625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.206282138824463, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117187240.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.345498499526065, + "step": 267, + "teaching_rms": 6.345503258154754, + "traffic_rms": 0.00018503892429886008 + }, + { + "batch_loss": 2.3359758853912354, + "innovation_rms": 6.131769623001766, + "instruction_rms": 6.131702812056304, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2615.132080078125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 710382592.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117292.8203125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7198.35888671875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.205567836761475, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117175512.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.5644673109054565, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.5139346942305565e-05, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117292.8203125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7198.3515625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.20556116104126, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117175512.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.131702812056304, + "step": 268, + "teaching_rms": 6.131769623001766, + "traffic_rms": 0.00018556591175933427 + }, + { + "batch_loss": 2.3295845985412598, + "innovation_rms": 6.631263099449841, + "instruction_rms": 6.6312473795435425, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2684.140625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 639344320.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117281.09375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7197.63818359375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.2048468589782715, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117163784.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.34454697370529175, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.3625412066176068e-05, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117281.09375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7197.630859375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.204840183258057, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117163784.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.6312473795435425, + "step": 269, + "teaching_rms": 6.631263099449841, + "traffic_rms": 0.00018586205789637293 + }, + { + "batch_loss": 2.301117181777954, + "innovation_rms": 6.227690101075762, + "instruction_rms": 6.227650446539298, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2664.101318359375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 575409856.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117269.359375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7196.91796875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.204125881195068, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117152056.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.09984084963798523, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.2262870768608991e-05, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117269.359375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7196.91064453125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.204118728637695, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117152056.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.227650446539298, + "step": 270, + "teaching_rms": 6.227690101075762, + "traffic_rms": 0.00018607769387552687 + }, + { + "batch_loss": 2.322810649871826, + "innovation_rms": 6.056624276410073, + "instruction_rms": 6.056589623760107, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2683.30419921875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 517868864.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117257.625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7196.19775390625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.203404903411865, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117140328.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.24830636382102966, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.1036583600798622e-05, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117257.625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7196.1904296875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.203397750854492, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117140328.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.056589651144562, + "step": 271, + "teaching_rms": 6.056624276410073, + "traffic_rms": 0.00018603934526988434 + }, + { + "batch_loss": 2.298269271850586, + "innovation_rms": 5.779281463416484, + "instruction_rms": 5.779248518724946, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2651.611572265625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 466081984.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117245.890625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7195.4775390625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.202683925628662, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117128600.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.38974282145500183, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 9.9329254226177e-06, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117245.890625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7195.47021484375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.202676773071289, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117128600.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 5.779248533074251, + "step": 272, + "teaching_rms": 5.779281463416484, + "traffic_rms": 0.00018557655290730388 + }, + { + "batch_loss": 2.330063581466675, + "innovation_rms": 6.172647000330377, + "instruction_rms": 6.172606333041623, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2600.721923828125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 419473792.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117234.15625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7194.75732421875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.201962947845459, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117116880.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.25596049427986145, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 8.93963260750752e-06, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117234.15625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7194.75, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.201955795288086, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117116880.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.172606333041623, + "step": 273, + "teaching_rms": 6.172647000330377, + "traffic_rms": 0.00018503919142657057 + }, + { + "batch_loss": 2.3256282806396484, + "innovation_rms": 6.534215474822455, + "instruction_rms": 6.53418195490701, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2618.8916015625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 377526400.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117222.421875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7194.037109375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.201242446899414, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117105160.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.17983432114124298, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 8.045669346756767e-06, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117222.421875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7194.02978515625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.201234817504883, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117105160.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.53418195490701, + "step": 274, + "teaching_rms": 6.534215474822455, + "traffic_rms": 0.0001844916916254183 + }, + { + "batch_loss": 2.3258776664733887, + "innovation_rms": 6.281332427432134, + "instruction_rms": 6.281297664956611, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2635.71044921875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 339773760.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117210.6953125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7193.31689453125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.200521945953369, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117093440.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2922641932964325, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.24110213923268e-06, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117210.6953125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7193.3095703125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.20051383972168, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117093440.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.281297664956611, + "step": 275, + "teaching_rms": 6.281332427432134, + "traffic_rms": 0.00018412002668095317 + }, + { + "batch_loss": 2.2988064289093018, + "innovation_rms": 6.159112411099266, + "instruction_rms": 6.159070363272192, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2647.8642578125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 305796384.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117198.96875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7192.5966796875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.199801445007324, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117081720.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.24254827201366425, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.5169915615115315e-06, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117198.96875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7192.58935546875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.199792861938477, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117081720.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.159070363272192, + "step": 276, + "teaching_rms": 6.159112411099266, + "traffic_rms": 0.0001838687232698029 + }, + { + "batch_loss": 2.3241872787475586, + "innovation_rms": 6.6492510911378755, + "instruction_rms": 6.649204271879292, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2665.98095703125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 275216736.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117187.2421875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7191.876953125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.199080944061279, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117070000.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.21931470930576324, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.865292223461438e-06, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117187.2421875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7191.86962890625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.199072360992432, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117070000.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.649204271879292, + "step": 277, + "teaching_rms": 6.6492510911378755, + "traffic_rms": 0.0001837915131005207 + }, + { + "batch_loss": 2.3610117435455322, + "innovation_rms": 6.968398230833062, + "instruction_rms": 6.968338098633251, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2656.986083984375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 247695056.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117175.515625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7191.1572265625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.198360443115234, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117058280.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2171192467212677, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.278762728266884e-06, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117175.515625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7191.14990234375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.198351860046387, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117058280.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.968338110533965, + "step": 278, + "teaching_rms": 6.968398230833062, + "traffic_rms": 0.00018383150026637677 + }, + { + "batch_loss": 2.38078236579895, + "innovation_rms": 6.260554877199166, + "instruction_rms": 6.260514680577034, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2646.7705078125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 222925552.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117163.7890625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7190.4375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.1976399421691895, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117046560.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.19133231043815613, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.7508865463896655e-06, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117163.7890625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7190.43017578125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.197631359100342, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117046560.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.260514680577034, + "step": 279, + "teaching_rms": 6.260554877199166, + "traffic_rms": 0.00018373363462322287 + }, + { + "batch_loss": 2.331296443939209, + "innovation_rms": 6.733733564844103, + "instruction_rms": 6.733702827582524, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2627.519775390625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 200632992.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117152.0625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7189.7177734375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.1969194412231445, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117034848.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3426187038421631, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.275797891750699e-06, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117152.0625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7189.71044921875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.196910858154297, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117034848.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.733702827582524, + "step": 280, + "teaching_rms": 6.733733564844103, + "traffic_rms": 0.000183182475633389 + }, + { + "batch_loss": 2.3390936851501465, + "innovation_rms": 6.235762150743483, + "instruction_rms": 6.235708800722366, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2724.747314453125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 180569696.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117140.3359375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7188.998046875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.1961989402771, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117023136.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4386723041534424, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.848218057100894e-06, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117140.3359375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7188.99072265625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.19619083404541, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117023136.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.235708800722366, + "step": 281, + "teaching_rms": 6.235762150743483, + "traffic_rms": 0.00018283499647675657 + }, + { + "batch_loss": 2.334242582321167, + "innovation_rms": 6.576739134352251, + "instruction_rms": 6.576671246191634, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2735.450439453125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 162512720.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117128.609375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7188.2783203125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.195478439331055, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117011424.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.26422351598739624, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.463396069491864e-06, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117128.609375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7188.27099609375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.195470809936523, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 117011424.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.576671246191634, + "step": 282, + "teaching_rms": 6.576739134352251, + "traffic_rms": 0.00018249112849936898 + }, + { + "batch_loss": 2.3252670764923096, + "innovation_rms": 6.545014576345903, + "instruction_rms": 6.5449785417466915, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2655.188720703125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 146261440.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117116.8828125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7187.55908203125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.19475793838501, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116999712.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.21039804816246033, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.117056394330575e-06, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117116.8828125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7187.5517578125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.194750785827637, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116999712.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.544978529076184, + "step": 283, + "teaching_rms": 6.545014576345903, + "traffic_rms": 0.00018243360287853387 + }, + { + "batch_loss": 2.343247890472412, + "innovation_rms": 6.43744830902761, + "instruction_rms": 6.437393830680152, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2732.619140625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 143646656.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117105.1640625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7186.83984375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.194037914276123, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116988000.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3506414592266083, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.80535073216015e-06, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117105.1640625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7186.83251953125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.19403076171875, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116988000.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.437393843562416, + "step": 284, + "teaching_rms": 6.43744830902761, + "traffic_rms": 0.00018266386726521682 + }, + { + "batch_loss": 2.336399793624878, + "innovation_rms": 6.0938985089303275, + "instruction_rms": 6.093880901986062, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2625.614013671875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 144687552.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117093.4453125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7186.12060546875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.1933183670043945, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116976288.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2830529808998108, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.5248154997825623e-06, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117093.4453125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7186.11328125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.193310737609863, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116976288.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.093880901986062, + "step": 285, + "teaching_rms": 6.0938985089303275, + "traffic_rms": 0.00018303972446984335 + }, + { + "batch_loss": 2.3309030532836914, + "innovation_rms": 6.177533150096755, + "instruction_rms": 6.177501431438714, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2563.67919921875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 145593872.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117081.734375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7185.4013671875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.192598342895508, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116964576.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3046990633010864, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.272333858854836e-06, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117081.734375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7185.39404296875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.192590713500977, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116964576.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.177501431438714, + "step": 286, + "teaching_rms": 6.177533150096755, + "traffic_rms": 0.00018350454425025275 + }, + { + "batch_loss": 2.3260931968688965, + "innovation_rms": 6.407042451430201, + "instruction_rms": 6.406996207036885, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2615.864013671875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 146281008.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117070.015625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7184.68212890625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.191878318786621, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116952872.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.34952273964881897, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.0451004729693523e-06, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117070.015625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7184.6748046875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.19187068939209, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116952872.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.406996207036885, + "step": 287, + "teaching_rms": 6.407042451430201, + "traffic_rms": 0.00018404243179179596 + }, + { + "batch_loss": 2.2892401218414307, + "innovation_rms": 6.731932072226192, + "instruction_rms": 6.731875332122377, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2646.594482421875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 145846864.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117058.296875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7183.962890625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.191158294677734, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116941168.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.17946496605873108, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.8405903574603144e-06, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117058.296875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7183.95556640625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.191150665283203, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116941168.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.731875344441114, + "step": 288, + "teaching_rms": 6.731932072226192, + "traffic_rms": 0.0001845587532713901 + }, + { + "batch_loss": 2.3030292987823486, + "innovation_rms": 6.545621740452894, + "instruction_rms": 6.545558407119962, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2689.61962890625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 145454448.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117046.578125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7183.24365234375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.190438270568848, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116929464.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.16820679605007172, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.6565312535021803e-06, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117046.578125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7183.236328125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.190431118011475, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116929464.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.545558419789348, + "step": 289, + "teaching_rms": 6.545621740452894, + "traffic_rms": 0.00018505212662449548 + }, + { + "batch_loss": 2.294813632965088, + "innovation_rms": 6.1487046442292845, + "instruction_rms": 6.148656835071432, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2753.789306640625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 145016272.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117034.859375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7182.52490234375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.189718246459961, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116917760.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11494273692369461, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.490878048571176e-06, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117034.859375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7182.517578125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.189711570739746, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116917760.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.148656835071432, + "step": 290, + "teaching_rms": 6.1487046442292845, + "traffic_rms": 0.00018533335883640656 + }, + { + "batch_loss": 2.3114736080169678, + "innovation_rms": 6.69798369697785, + "instruction_rms": 6.697963428159754, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2788.370849609375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 144255200.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117023.140625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7181.80615234375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.188998699188232, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116906056.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.150255486369133, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.3417901527645881e-06, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117023.140625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7181.798828125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.188992023468018, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116906056.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.697963428159754, + "step": 291, + "teaching_rms": 6.69798369697785, + "traffic_rms": 0.00018520690489361345 + }, + { + "batch_loss": 2.323643445968628, + "innovation_rms": 6.478099201072765, + "instruction_rms": 6.478094669801708, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2770.1181640625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 144464640.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117011.421875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7181.08740234375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.188279151916504, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116894352.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2482614368200302, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.207611148856813e-06, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 117011.421875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7181.080078125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.188272476196289, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116894352.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.478094669801708, + "step": 292, + "teaching_rms": 6.478099201072765, + "traffic_rms": 0.0001851100868447983 + }, + { + "batch_loss": 2.3325355052948, + "innovation_rms": 6.4906590186707955, + "instruction_rms": 6.4906785927554544, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2782.699462890625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 144174416.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116999.7109375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7180.36865234375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.187559604644775, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116882656.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4092676043510437, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.0868500339711318e-06, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116999.7109375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7180.361328125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.187553405761719, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116882656.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.490678605531961, + "step": 293, + "teaching_rms": 6.4906590186707955, + "traffic_rms": 0.0001849476909850132 + }, + { + "batch_loss": 2.3259315490722656, + "innovation_rms": 6.896209899405662, + "instruction_rms": 6.896180701315592, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2830.95947265625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 142720608.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116988.0, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7179.64990234375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.186840057373047, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116870960.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2768603563308716, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 9.78164962361916e-07, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116988.0, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7179.642578125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.18683385848999, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116870960.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.896180701315592, + "step": 294, + "teaching_rms": 6.896209899405662, + "traffic_rms": 0.00018498342015245466 + }, + { + "batch_loss": 2.3205902576446533, + "innovation_rms": 6.283666568635043, + "instruction_rms": 6.283618347293503, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2838.888427734375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 143827040.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116976.2890625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7178.93115234375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.186120986938477, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116859264.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.12692809104919434, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 8.803484661257244e-07, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116976.2890625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7178.923828125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.186114311218262, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116859264.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.283618347293503, + "step": 295, + "teaching_rms": 6.283666568635043, + "traffic_rms": 0.0001850675982575969 + }, + { + "batch_loss": 2.3109331130981445, + "innovation_rms": 6.361617124296339, + "instruction_rms": 6.3615916514837, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2878.19677734375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 143472544.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116964.578125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7178.21240234375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.185401916503906, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116847568.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.24292975664138794, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.923135854071006e-07, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116964.578125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7178.205078125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.185394763946533, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116847568.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.361591664519462, + "step": 296, + "teaching_rms": 6.361617124296339, + "traffic_rms": 0.0001849637058276959 + }, + { + "batch_loss": 2.3413121700286865, + "innovation_rms": 5.988715203027085, + "instruction_rms": 5.988639237374529, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2851.441162109375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 143829424.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116952.875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7177.494140625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.184682846069336, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116835872.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.5254322290420532, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.130822154977068e-07, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116952.875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7177.48681640625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.184675693511963, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116835872.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 5.988639237374529, + "step": 297, + "teaching_rms": 5.988715203027085, + "traffic_rms": 0.00018484894558179753 + }, + { + "batch_loss": 2.299436569213867, + "innovation_rms": 6.381788201402707, + "instruction_rms": 6.381759353769093, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2814.921142578125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 145237616.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116941.171875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7176.77587890625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.183963775634766, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116824176.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.5286310911178589, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.417739655262267e-07, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116941.171875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7176.7685546875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.183956623077393, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116824176.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.381759353769093, + "step": 298, + "teaching_rms": 6.381788201402707, + "traffic_rms": 0.00018447449704220086 + }, + { + "batch_loss": 2.346522808074951, + "innovation_rms": 6.740169649841865, + "instruction_rms": 6.740144096112867, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2792.87451171875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 145148752.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116929.46875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7176.0576171875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.183244705200195, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116812480.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.24813365936279297, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.775965519205783e-07, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116929.46875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7176.05029296875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.183237552642822, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116812480.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.740144108416492, + "step": 299, + "teaching_rms": 6.740169649841865, + "traffic_rms": 0.00018436187208056522 + }, + { + "batch_loss": 2.348090887069702, + "innovation_rms": 6.11466636311055, + "instruction_rms": 6.114681351407388, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2750.519775390625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 144521520.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116917.765625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7175.33935546875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.182525634765625, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116800792.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.29667702317237854, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.19836873991153e-07, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116917.765625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7175.33203125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.18251895904541, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116800792.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.114681351407388, + "step": 300, + "teaching_rms": 6.11466636311055, + "traffic_rms": 0.00018434817120636447 + }, + { + "batch_loss": 2.327596664428711, + "innovation_rms": 6.366644247859935, + "instruction_rms": 6.366600965876852, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2697.64501953125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 144357360.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116906.0625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7174.62109375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.181807041168213, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116789104.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.37959352135658264, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.67853169539012e-07, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116906.0625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7174.61376953125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.18179988861084, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116789104.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.366600965876852, + "step": 301, + "teaching_rms": 6.366644247859935, + "traffic_rms": 0.00018419846101159796 + }, + { + "batch_loss": 2.3213629722595215, + "innovation_rms": 6.465648772117037, + "instruction_rms": 6.4655954178323105, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2605.37646484375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 144606752.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116894.359375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7173.90283203125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.181088447570801, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116777416.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.6371521353721619, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.210678525851108e-07, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116894.359375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7173.8955078125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.1810808181762695, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116777416.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.4655954178323105, + "step": 302, + "teaching_rms": 6.465648772117037, + "traffic_rms": 0.00018403753024628504 + }, + { + "batch_loss": 2.337709665298462, + "innovation_rms": 6.448871983431422, + "instruction_rms": 6.448792089968623, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2586.149658203125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 145233696.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116882.6640625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7173.1845703125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.1803693771362305, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116765728.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.6531166434288025, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.7896106164225785e-07, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116882.6640625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7173.17724609375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.180362224578857, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116765728.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.448792089968623, + "step": 303, + "teaching_rms": 6.448871983431422, + "traffic_rms": 0.00018386149837554047 + }, + { + "batch_loss": 2.3156936168670654, + "innovation_rms": 6.584645014559821, + "instruction_rms": 6.5845969169234815, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2587.82666015625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 144894832.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116870.96875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7172.466796875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.179650783538818, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116754040.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3146752119064331, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.4106494695151923e-07, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116870.96875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7172.45947265625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.179643630981445, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116754040.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.584596929517753, + "step": 304, + "teaching_rms": 6.584645014559821, + "traffic_rms": 0.00018368120076853845 + }, + { + "batch_loss": 2.310894250869751, + "innovation_rms": 6.344575615656818, + "instruction_rms": 6.344503440179632, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2575.609130859375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 144597984.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116859.2734375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7171.7490234375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.178932189941406, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116742352.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.30503374338150024, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.0695844088768354e-07, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116859.2734375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7171.74169921875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.178925037384033, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116742352.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.344503440179632, + "step": 305, + "teaching_rms": 6.344575615656818, + "traffic_rms": 0.00018374029288733588 + }, + { + "batch_loss": 2.3276708126068115, + "innovation_rms": 6.15570506715524, + "instruction_rms": 6.15565058779062, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2593.840576171875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 144401632.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116847.578125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7171.03125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.178213596343994, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116730664.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3796680271625519, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.762625967989152e-07, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116847.578125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7171.02392578125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.178206443786621, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116730664.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.15565058779062, + "step": 306, + "teaching_rms": 6.15570506715524, + "traffic_rms": 0.0001838344108956026 + }, + { + "batch_loss": 2.327651262283325, + "innovation_rms": 6.759254123176603, + "instruction_rms": 6.759213734275597, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2594.09130859375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 145045168.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116835.890625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7170.3134765625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.177495002746582, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116718984.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.5988670587539673, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.486363257503399e-07, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116835.890625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7170.30615234375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.177488327026367, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116718984.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.759213734275597, + "step": 307, + "teaching_rms": 6.759254123176603, + "traffic_rms": 0.00018392080131831326 + }, + { + "batch_loss": 2.3277695178985596, + "innovation_rms": 6.80147884292368, + "instruction_rms": 6.801440449833168, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2563.060546875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 146928672.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116824.1953125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7169.595703125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.176776885986328, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116707304.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.540228009223938, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.2377268749096402e-07, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116824.1953125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7169.58837890625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.176769733428955, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116707304.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.801440449833168, + "step": 308, + "teaching_rms": 6.80147884292368, + "traffic_rms": 0.0001841879616454497 + }, + { + "batch_loss": 2.3205454349517822, + "innovation_rms": 6.332342850944266, + "instruction_rms": 6.33229422841313, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2584.722412109375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 147177344.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116812.5, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7168.8779296875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.176058292388916, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116695624.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2735595703125, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.0139540879426931e-07, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116812.5, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7168.87060546875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.176051616668701, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116695624.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.33229422841313, + "step": 309, + "teaching_rms": 6.332342850944266, + "traffic_rms": 0.00018435293043170422 + }, + { + "batch_loss": 2.33591890335083, + "innovation_rms": 6.307061193430264, + "instruction_rms": 6.307035753210692, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2666.6572265625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 146462976.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116800.8125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7168.16015625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.175340175628662, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116683944.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4854196310043335, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.812558565461586e-07, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116800.8125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7168.15283203125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.175333499908447, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116683944.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.307035753210692, + "step": 310, + "teaching_rms": 6.307061193430264, + "traffic_rms": 0.0001842310823801205 + }, + { + "batch_loss": 2.3357276916503906, + "innovation_rms": 6.066072702640255, + "instruction_rms": 6.0660237917106565, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2658.528076171875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 146894720.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116789.125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7167.44287109375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.174622058868408, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116672264.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.5296362042427063, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.6313026662828634e-07, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116789.125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7167.435546875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.174615383148193, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116672264.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.0660237917106565, + "step": 311, + "teaching_rms": 6.066072702640255, + "traffic_rms": 0.00018406650789938448 + }, + { + "batch_loss": 2.3259575366973877, + "innovation_rms": 6.301446072885848, + "instruction_rms": 6.301389998877578, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2690.56396484375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 146878848.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116777.4375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7166.7255859375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.173903942108154, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116660584.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.34781694412231445, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.4681724280762865e-07, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116777.4375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7166.71826171875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.1738972663879395, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116660584.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.301389998877578, + "step": 312, + "teaching_rms": 6.301446072885848, + "traffic_rms": 0.00018381602543139378 + }, + { + "batch_loss": 2.344186544418335, + "innovation_rms": 6.813966994186366, + "instruction_rms": 6.813930663699016, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2741.58349609375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 145433104.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116765.7421875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7166.00830078125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.1731858253479, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116648904.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.26324132084846497, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.3213551142143842e-07, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116765.7421875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7166.0009765625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.1731791496276855, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116648904.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.813930663699016, + "step": 313, + "teaching_rms": 6.813966994186366, + "traffic_rms": 0.00018367509876928507 + }, + { + "batch_loss": 2.318521022796631, + "innovation_rms": 5.904570979765386, + "instruction_rms": 5.904497878792977, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2792.561767578125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 144167088.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116754.0546875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7165.291015625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.172468185424805, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116637232.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.42070573568344116, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.1892196027929458e-07, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116754.0546875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7165.28369140625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.172461032867432, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116637232.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 5.904497878792977, + "step": 314, + "teaching_rms": 5.904570979765386, + "traffic_rms": 0.00018382627042878716 + }, + { + "batch_loss": 2.325622320175171, + "innovation_rms": 6.104456355435098, + "instruction_rms": 6.104446913986636, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2747.30517578125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 143979808.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116742.375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7164.57373046875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.171750068664551, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116625560.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3886701762676239, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.0702976283027965e-07, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116742.375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7164.56640625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.171743392944336, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116625560.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.10444692757152, + "step": 315, + "teaching_rms": 6.104456355435098, + "traffic_rms": 0.00018378450705584294 + }, + { + "batch_loss": 2.3041412830352783, + "innovation_rms": 6.355309626503075, + "instruction_rms": 6.355277266227254, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2820.94775390625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 143613648.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116730.6875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7163.8564453125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.171032428741455, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116613888.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.41272062063217163, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 9.632678654725169e-08, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116730.6875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7163.84912109375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.171025276184082, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116613888.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.355277279275969, + "step": 316, + "teaching_rms": 6.355309626503075, + "traffic_rms": 0.00018396048497002206 + }, + { + "batch_loss": 2.325312852859497, + "innovation_rms": 6.545011991807075, + "instruction_rms": 6.544996875457787, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2833.962158203125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 143479344.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116719.0078125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7163.1396484375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.170314788818359, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116602216.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2658461928367615, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 8.669410789252652e-08, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116719.0078125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7163.13232421875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.170307636260986, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116602216.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.544996875457787, + "step": 317, + "teaching_rms": 6.545011991807075, + "traffic_rms": 0.0001841759788899151 + }, + { + "batch_loss": 2.303226947784424, + "innovation_rms": 6.479358639604041, + "instruction_rms": 6.479330866215064, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2819.48583984375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 142491056.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116707.3203125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7162.4228515625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.169597148895264, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116590544.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2650202512741089, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.80246978138166e-08, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116707.3203125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7162.41552734375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.169589996337891, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116590544.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.479330879013948, + "step": 318, + "teaching_rms": 6.479358639604041, + "traffic_rms": 0.00018458602901787335 + }, + { + "batch_loss": 2.3050684928894043, + "innovation_rms": 6.475160440121357, + "instruction_rms": 6.475133738372171, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2901.937744140625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 141781136.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116695.640625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7161.7060546875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.168879508972168, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116578872.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3020574748516083, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.022222803243494e-08, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116695.640625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7161.69873046875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.168872356414795, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116578872.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.47513376398653, + "step": 319, + "teaching_rms": 6.475160440121357, + "traffic_rms": 0.00018499609206042457 + }, + { + "batch_loss": 2.298576831817627, + "innovation_rms": 6.678851216126064, + "instruction_rms": 6.6788278107763075, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2904.374755859375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 141300752.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116683.953125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7160.9892578125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.1681623458862305, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116567200.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.34833449125289917, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.320000522919145e-08, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116683.953125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7160.98193359375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.168154716491699, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116567200.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.678827823192887, + "step": 320, + "teaching_rms": 6.678851216126064, + "traffic_rms": 0.00018528513923406 + }, + { + "batch_loss": 2.350660800933838, + "innovation_rms": 7.066900199936269, + "instruction_rms": 7.066853519076067, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2947.692626953125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 140994800.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116672.2734375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7160.2724609375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.167444705963135, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116555536.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3002622127532959, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.68800047062723e-08, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116672.2734375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7160.26513671875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.167437553405762, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116555536.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 7.066853530810879, + "step": 321, + "teaching_rms": 7.066900199936269, + "traffic_rms": 0.00018554835719289186 + }, + { + "batch_loss": 2.3076624870300293, + "innovation_rms": 6.4692145621922945, + "instruction_rms": 6.469177939423824, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2933.1748046875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 141080720.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116660.6015625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7159.5556640625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.166727542877197, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116543872.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.31256192922592163, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.1192003525102336e-08, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116660.6015625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7159.54833984375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.166719913482666, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116543872.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.469177939423824, + "step": 322, + "teaching_rms": 6.4692145621922945, + "traffic_rms": 0.0001856964733500433 + }, + { + "batch_loss": 2.339111089706421, + "innovation_rms": 6.913233960284256, + "instruction_rms": 6.913166293449238, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2968.48486328125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 141001936.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116648.921875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7158.8388671875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.16601037979126, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116532208.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.19155248999595642, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.607280246204937e-08, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116648.921875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7158.83154296875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.1660027503967285, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116532208.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.913166293449238, + "step": 323, + "teaching_rms": 6.913233960284256, + "traffic_rms": 0.0001856071677759874 + }, + { + "batch_loss": 2.3245480060577393, + "innovation_rms": 6.35560073220243, + "instruction_rms": 6.355521233130861, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 3006.855224609375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 141421568.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116637.2421875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7158.12255859375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.165292739868164, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116520544.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.36372435092926025, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.14655225711158e-08, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116637.2421875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7158.115234375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.165285587310791, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116520544.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.355521246179075, + "step": 324, + "teaching_rms": 6.35560073220243, + "traffic_rms": 0.0001851857530002812 + }, + { + "batch_loss": 2.287607192993164, + "innovation_rms": 6.564856332573171, + "instruction_rms": 6.5648314731548325, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2951.76806640625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 140424128.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116625.5703125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7157.40625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.164575576782227, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116508880.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2889968454837799, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.7318969248190115e-08, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116625.5703125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7157.39892578125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.164567947387695, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116508880.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.5648314731548325, + "step": 325, + "teaching_rms": 6.564856332573171, + "traffic_rms": 0.00018435865474561696 + }, + { + "batch_loss": 2.3452048301696777, + "innovation_rms": 5.773760253950098, + "instruction_rms": 5.773735913378582, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2934.713623046875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 140069456.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116613.8984375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7156.68994140625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.163858413696289, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116497216.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.21735242009162903, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.358707090228563e-08, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116613.8984375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7156.6826171875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.163850784301758, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116497216.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 5.773735913378582, + "step": 326, + "teaching_rms": 5.773760253950098, + "traffic_rms": 0.0001837211043264552 + }, + { + "batch_loss": 2.3138463497161865, + "innovation_rms": 6.611616518350881, + "instruction_rms": 6.611531128124851, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2969.645263671875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 139245168.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116602.2265625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7155.9736328125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.16314172744751, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116485552.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.34114205837249756, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.022836381205707e-08, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116602.2265625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7155.96630859375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.16313362121582, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116485552.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.611531128124851, + "step": 327, + "teaching_rms": 6.611616518350881, + "traffic_rms": 0.00018290788802044955 + }, + { + "batch_loss": 2.324497938156128, + "innovation_rms": 6.479210374004736, + "instruction_rms": 6.479192597889245, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2881.775634765625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 139984768.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116590.5546875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7155.25732421875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.162424564361572, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116473896.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.34363433718681335, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.7205526720308626e-08, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116590.5546875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7155.25, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.162416934967041, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116473896.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.479192597889245, + "step": 328, + "teaching_rms": 6.479210374004736, + "traffic_rms": 0.0001821935522459808 + }, + { + "batch_loss": 2.3420169353485107, + "innovation_rms": 6.356045682669413, + "instruction_rms": 6.356043909470828, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2820.898193359375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 140687376.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116578.890625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7154.541015625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.161707401275635, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116462240.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.42481690645217896, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.4484974048277763e-08, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116578.890625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7154.53369140625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.161700248718262, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116462240.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.356043909470828, + "step": 329, + "teaching_rms": 6.356045682669413, + "traffic_rms": 0.00018192870007981723 + }, + { + "batch_loss": 2.300502061843872, + "innovation_rms": 6.558503788813574, + "instruction_rms": 6.5584482420372945, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2790.840576171875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 141233904.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116567.2265625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7153.82470703125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.1609907150268555, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116450584.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.26789748668670654, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.2036475755271567e-08, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116567.2265625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7153.8173828125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.160983085632324, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116450584.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.5584482420372945, + "step": 330, + "teaching_rms": 6.558503788813574, + "traffic_rms": 0.00018233955473301346 + }, + { + "batch_loss": 2.3220653533935547, + "innovation_rms": 6.604433969730207, + "instruction_rms": 6.604379613266801, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2848.843017578125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 140549472.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116555.5625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7153.10888671875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.160274028778076, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116438928.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.22979636490345, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.9832826936294623e-08, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116555.5625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7153.1015625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.160266399383545, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116438928.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.604379625823348, + "step": 331, + "teaching_rms": 6.604433969730207, + "traffic_rms": 0.0001831618388324071 + }, + { + "batch_loss": 2.3120439052581787, + "innovation_rms": 6.088726755817639, + "instruction_rms": 6.0886654550706165, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2876.693603515625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 140614432.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116543.890625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7152.39306640625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.159557342529297, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116427272.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2349226027727127, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.7849544065029477e-08, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116543.890625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7152.3857421875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.159549713134766, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116427272.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.0886654550706165, + "step": 332, + "teaching_rms": 6.088726755817639, + "traffic_rms": 0.00018411420769524867 + }, + { + "batch_loss": 2.32232928276062, + "innovation_rms": 6.152706045646764, + "instruction_rms": 6.152674345494996, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2949.6015625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 139740832.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116532.2265625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7151.67724609375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.158841133117676, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116415616.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.29092228412628174, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.606458965852653e-08, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116532.2265625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7151.669921875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.158833026885986, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116415616.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.152674358973396, + "step": 333, + "teaching_rms": 6.152706045646764, + "traffic_rms": 0.00018478275162488043 + }, + { + "batch_loss": 2.323047637939453, + "innovation_rms": 6.479721628842034, + "instruction_rms": 6.479691835516988, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 3000.06201171875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 139874944.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116520.5625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7150.96142578125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.1581244468688965, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116403960.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.20708055794239044, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.4458130692673876e-08, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116520.5625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7150.9541015625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.158116817474365, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116403960.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.479691835516988, + "step": 334, + "teaching_rms": 6.479721628842034, + "traffic_rms": 0.00018526875865194288 + }, + { + "batch_loss": 2.3226511478424072, + "innovation_rms": 6.285045207581321, + "instruction_rms": 6.285033095815178, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 3026.640380859375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 140359296.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116508.8984375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7150.24560546875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.157407760620117, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116392312.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.19446714222431183, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.3012317445770805e-08, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116508.8984375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7150.23828125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.157400608062744, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116392312.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.285033095815178, + "step": 335, + "teaching_rms": 6.285045207581321, + "traffic_rms": 0.00018560060977438757 + }, + { + "batch_loss": 2.298741579055786, + "innovation_rms": 5.743622772566537, + "instruction_rms": 5.743583329278729, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2981.072998046875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 141105984.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116497.234375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7149.52978515625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.156691074371338, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116380664.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.22454802691936493, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.1711085434740198e-08, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116497.234375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7149.5224609375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.156684398651123, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116380664.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 5.743583343717137, + "step": 336, + "teaching_rms": 5.743622772566537, + "traffic_rms": 0.0001857184986048867 + }, + { + "batch_loss": 2.3137261867523193, + "innovation_rms": 6.343408724354852, + "instruction_rms": 6.3433961480240395, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2967.216064453125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 140432784.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116485.578125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7148.81396484375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.155974864959717, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116369016.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.22917652130126953, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 1.0539976891266178e-08, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116485.578125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7148.806640625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.155967712402344, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116369016.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.3433961480240395, + "step": 337, + "teaching_rms": 6.343408724354852, + "traffic_rms": 0.00018553586953098136 + }, + { + "batch_loss": 2.2894296646118164, + "innovation_rms": 5.936102060565386, + "instruction_rms": 5.936020922567239, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2836.674560546875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 141518896.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116473.9140625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7148.0986328125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.155258655548096, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116357368.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14356592297554016, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 9.485979113321719e-09, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116473.9140625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7148.09130859375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.1552510261535645, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116357368.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 5.936020922567239, + "step": 338, + "teaching_rms": 5.936102060565386, + "traffic_rms": 0.00018532520419676498 + }, + { + "batch_loss": 2.323439121246338, + "innovation_rms": 6.588244718726057, + "instruction_rms": 6.588213476266131, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2801.874755859375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 142369840.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116462.2578125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7147.38330078125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.154542446136475, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116345720.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.20378348231315613, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 8.537380580264653e-09, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116462.2578125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7147.3759765625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.154534816741943, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116345720.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.588213476266131, + "step": 339, + "teaching_rms": 6.588244718726057, + "traffic_rms": 0.0001847881817592601 + }, + { + "batch_loss": 2.318883180618286, + "innovation_rms": 6.287715926368904, + "instruction_rms": 6.287673141137641, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2801.969970703125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 142508960.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116450.6015625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7146.66796875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.1538262367248535, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116334072.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.28921782970428467, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.68364216696682e-09, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116450.6015625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7146.66064453125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.153818607330322, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116334072.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.287673141137641, + "step": 340, + "teaching_rms": 6.287715926368904, + "traffic_rms": 0.00018432387713337902 + }, + { + "batch_loss": 2.307809829711914, + "innovation_rms": 6.74076766634721, + "instruction_rms": 6.740739109583374, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2776.56591796875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 142563072.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116438.9453125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7145.95263671875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.153110504150391, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116322432.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3145782947540283, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.915277683816612e-09, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116438.9453125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7145.9453125, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.153102397918701, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116322432.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.740739097280835, + "step": 341, + "teaching_rms": 6.74076766634721, + "traffic_rms": 0.00018395899371284095 + }, + { + "batch_loss": 2.314749002456665, + "innovation_rms": 6.486690780641876, + "instruction_rms": 6.48667240867232, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2729.28662109375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 143414384.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116427.2890625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7145.2373046875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.152394771575928, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116310792.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.35417720675468445, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 6.223749959843872e-09, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116427.2890625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7145.22998046875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.15238618850708, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116310792.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.48667240867232, + "step": 342, + "teaching_rms": 6.486690780641876, + "traffic_rms": 0.00018417770992965163 + }, + { + "batch_loss": 2.3250651359558105, + "innovation_rms": 6.358091888104337, + "instruction_rms": 6.358069090306716, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2805.83740234375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 142752896.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116415.6328125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7144.52197265625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.151679039001465, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116299152.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2345060110092163, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.6013749194505635e-09, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116415.6328125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7144.5146484375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.151670455932617, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116299152.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.358069090306716, + "step": 343, + "teaching_rms": 6.358091888104337, + "traffic_rms": 0.00018455758154249828 + }, + { + "batch_loss": 2.314680814743042, + "innovation_rms": 6.543731792883309, + "instruction_rms": 6.543686122140339, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2881.500732421875, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 142269072.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116403.984375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7143.806640625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.150963306427002, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116287512.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.2810690701007843, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 5.041237205460902e-09, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116403.984375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7143.79931640625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.150954723358154, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116287512.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.543686122140339, + "step": 344, + "teaching_rms": 6.543731792883309, + "traffic_rms": 0.00018484695342285234 + }, + { + "batch_loss": 2.3138134479522705, + "innovation_rms": 6.26735784601638, + "instruction_rms": 6.267305331729926, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2846.5517578125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 142115488.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116392.3359375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7143.091796875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.150247573852539, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116275872.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3674173951148987, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.537113351688049e-09, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116392.3359375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7143.08447265625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.150238990783691, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116275872.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.267305331729926, + "step": 345, + "teaching_rms": 6.26735784601638, + "traffic_rms": 0.0001847824573765879 + }, + { + "batch_loss": 2.2761194705963135, + "innovation_rms": 6.860400877303943, + "instruction_rms": 6.860353866735314, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2808.802734375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 142215104.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116380.6875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7142.376953125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.149531841278076, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116264232.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3311331570148468, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 4.083402060928165e-09, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116380.6875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7142.36962890625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.1495232582092285, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116264232.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.860353866735314, + "step": 346, + "teaching_rms": 6.860400877303943, + "traffic_rms": 0.00018449964198749075 + }, + { + "batch_loss": 2.3424220085144043, + "innovation_rms": 6.364358942199689, + "instruction_rms": 6.364282627927493, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2879.73876953125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 142010640.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116369.0390625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7141.662109375, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.148816108703613, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116252592.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3355037271976471, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.6750618104264277e-09, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116369.0390625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7141.65478515625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.148807525634766, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116252592.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.364282627927493, + "step": 347, + "teaching_rms": 6.364358942199689, + "traffic_rms": 0.00018436253564891404 + }, + { + "batch_loss": 2.3175699710845947, + "innovation_rms": 6.849707228627267, + "instruction_rms": 6.849647880420544, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2861.4619140625, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 143031168.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116357.390625, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7140.947265625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.14810037612915, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116240960.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.4275456368923187, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 3.3075555627704034e-09, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116357.390625, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7140.93994140625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.148092269897461, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116240960.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.849647880420544, + "step": 348, + "teaching_rms": 6.849707228627267, + "traffic_rms": 0.00018383896160229518 + }, + { + "batch_loss": 2.2963449954986572, + "innovation_rms": 6.328822897850293, + "instruction_rms": 6.3287735203287205, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2850.9892578125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 142596176.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116345.7421875, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7140.232421875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.1473846435546875, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116229328.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.3679439425468445, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.976799917675521e-09, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116345.7421875, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7140.22509765625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.147377014160156, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116229328.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.3287735203287205, + "step": 349, + "teaching_rms": 6.328822897850293, + "traffic_rms": 0.00018356787367988456 + }, + { + "batch_loss": 2.318504571914673, + "innovation_rms": 6.247094727554016, + "instruction_rms": 6.247075640835653, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2807.304443359375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 143035200.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116334.09375, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7139.517578125, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.146668910980225, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116217696.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.12766197323799133, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.6791198148856665e-09, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116334.09375, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7139.51025390625, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.146661758422852, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116217696.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.247075640835653, + "step": 350, + "teaching_rms": 6.247094727554016, + "traffic_rms": 0.00018333007667527357 + }, + { + "batch_loss": 2.323054075241089, + "innovation_rms": 5.961981093453792, + "instruction_rms": 5.961940384054717, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2763.046630859375, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 142878528.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116322.453125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7138.80322265625, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.14595365524292, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116206064.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.169703871011734, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.4112076779658764e-09, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116322.453125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7138.7958984375, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.145946502685547, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116206064.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 5.961940384054717, + "step": 351, + "teaching_rms": 5.961981093453792, + "traffic_rms": 0.0001831780300463299 + }, + { + "batch_loss": 2.327579975128174, + "innovation_rms": 6.235399441812911, + "instruction_rms": 6.235367485130701, + "parameter_state": { + "bn_running_mean": { + "all_finite": true, + "max_abs_over_finite_tensors": 2841.102783203125, + "nonfinite_indices": [] + }, + "bn_running_var": { + "all_finite": true, + "max_abs_over_finite_tensors": 143663712.0, + "nonfinite_indices": [] + }, + "feedback_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116310.8125, + "nonfinite_indices": [] + }, + "feedback_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7138.0888671875, + "nonfinite_indices": [] + }, + "feedback_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.145238876342773, + "nonfinite_indices": [] + }, + "feedback_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116194432.0, + "nonfinite_indices": [] + }, + "forward_bn_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 494176.78125, + "nonfinite_indices": [] + }, + "forward_bn_bias_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.40935218334198, + "nonfinite_indices": [] + }, + "forward_bn_scale": { + "all_finite": true, + "max_abs_over_finite_tensors": 649938.0625, + "nonfinite_indices": [] + }, + "forward_bn_scale_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 2.170086776942526e-09, + "nonfinite_indices": [] + }, + "forward_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 116310.8125, + "nonfinite_indices": [] + }, + "forward_readout": { + "all_finite": true, + "max_abs_over_finite_tensors": 7138.08154296875, + "nonfinite_indices": [] + }, + "forward_readout_momentum": { + "all_finite": true, + "max_abs_over_finite_tensors": 7.145231246948242, + "nonfinite_indices": [] + }, + "forward_weight": { + "all_finite": true, + "max_abs_over_finite_tensors": 116194432.0, + "nonfinite_indices": [] + }, + "predictor_bias": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.11661697179079056, + "nonfinite_indices": [] + }, + "predictor_slope": { + "all_finite": true, + "max_abs_over_finite_tensors": 0.14589270949363708, + "nonfinite_indices": [] + } + }, + "predictor_mse": null, + "predictor_updated": false, + "raw_apical_rms": 6.235367485130701, + "step": 352, + "teaching_rms": 6.235399441812911, + "traffic_rms": 0.00018287902183983132 + } + ], + "trajectory_metrics_finite": true, + "validation_evaluations": 0 +} -- cgit v1.2.3