summaryrefslogtreecommitdiff
path: root/results/kp_dynamic_projection_confirmation/seed11_clean_kp.json
diff options
context:
space:
mode:
authorYurenHao0426 <Blackhao0426@gmail.com>2026-07-23 06:52:10 -0500
committerYurenHao0426 <Blackhao0426@gmail.com>2026-07-23 06:52:10 -0500
commit2fed62a2486962e0054fbb9d38661ed76a2820f4 (patch)
tree5e94f7e4e2319e6fd7045e095e9805a0a9cf4019 /results/kp_dynamic_projection_confirmation/seed11_clean_kp.json
parentb3beab598f062ac0bca2a27b6b62b790223366e5 (diff)
results: pass dynamic innovation confirmation gate
Diffstat (limited to 'results/kp_dynamic_projection_confirmation/seed11_clean_kp.json')
-rw-r--r--results/kp_dynamic_projection_confirmation/seed11_clean_kp.json3258
1 files changed, 3258 insertions, 0 deletions
diff --git a/results/kp_dynamic_projection_confirmation/seed11_clean_kp.json b/results/kp_dynamic_projection_confirmation/seed11_clean_kp.json
new file mode 100644
index 0000000..e3f64f1
--- /dev/null
+++ b/results/kp_dynamic_projection_confirmation/seed11_clean_kp.json
@@ -0,0 +1,3258 @@
+{
+ "architecture": {
+ "adaptive_apical_parameters": 0,
+ "adaptive_feedback_parameters": 267904,
+ "base_width": 16,
+ "blocks_per_stage": 3,
+ "bn_eps": 1e-05,
+ "bn_momentum": 0.1,
+ "depth": 20,
+ "family": "CIFAR 6n+2 ResNet, option-A shortcuts",
+ "fixed_feedback_parameters": 0,
+ "fixed_traffic_coefficients": 0,
+ "forward_parameters": 269722,
+ "hidden_shapes": [
+ [
+ 16,
+ 32,
+ 32
+ ],
+ [
+ 16,
+ 32,
+ 32
+ ],
+ [
+ 16,
+ 32,
+ 32
+ ],
+ [
+ 16,
+ 32,
+ 32
+ ],
+ [
+ 16,
+ 32,
+ 32
+ ],
+ [
+ 16,
+ 32,
+ 32
+ ],
+ [
+ 16,
+ 32,
+ 32
+ ],
+ [
+ 32,
+ 16,
+ 16
+ ],
+ [
+ 32,
+ 16,
+ 16
+ ],
+ [
+ 32,
+ 16,
+ 16
+ ],
+ [
+ 32,
+ 16,
+ 16
+ ],
+ [
+ 32,
+ 16,
+ 16
+ ],
+ [
+ 32,
+ 16,
+ 16
+ ],
+ [
+ 64,
+ 8,
+ 8
+ ],
+ [
+ 64,
+ 8,
+ 8
+ ],
+ [
+ 64,
+ 8,
+ 8
+ ],
+ [
+ 64,
+ 8,
+ 8
+ ],
+ [
+ 64,
+ 8,
+ 8
+ ],
+ [
+ 64,
+ 8,
+ 8
+ ]
+ ],
+ "normalization": "batchnorm",
+ "predictor_parameters": 0,
+ "residual_scale": 1.0,
+ "vectorizer_mode": null,
+ "vectorizer_parameters": 0
+ },
+ "args": {
+ "a_scale": 1.0,
+ "a_warmup_steps": 0,
+ "alignment_probe": 32,
+ "apical_calibration_mode": "unit_targets",
+ "apical_seed": null,
+ "augment_train": 1,
+ "batch_size": 128,
+ "bn_eps": 1e-05,
+ "bn_momentum": 0.1,
+ "data_dir": "/home/yurenh2/sdrn/data",
+ "depth": 20,
+ "device": "cuda",
+ "epochs": 200,
+ "eta_A": 0.01,
+ "eta_P": 0.01,
+ "eval_every": 0,
+ "eval_split": "test",
+ "learn_P": 0,
+ "loader_seed": 11,
+ "lr": 0.1,
+ "lr_gamma": 0.1,
+ "lr_milestones": "100,150",
+ "lr_schedule": "step",
+ "max_steps": 0,
+ "mirror_batch_size": 1,
+ "mirror_eta": 0.1,
+ "mirror_every": 16,
+ "mirror_noise_std": 1.0,
+ "mirror_seed": 3000,
+ "mirror_warmup_steps": 0,
+ "mode": "kp",
+ "momentum": 0.9,
+ "neutral_projection": 0,
+ "normalization": "batchnorm",
+ "nuisance_scale": 0.0,
+ "out": "results/kp_dynamic_projection_confirmation/seed11_clean_kp.json",
+ "output_lr": 0.1,
+ "pert_directions": 1,
+ "pert_every": 4,
+ "pert_sigma": 0.01,
+ "perturb_seed": 1000,
+ "predictor_every": 0,
+ "predictor_mode": "nlms",
+ "predictor_warmup_steps": 0,
+ "residual_scale": null,
+ "seed": 11,
+ "split_seed": 2027,
+ "traffic_calibration_examples": 64,
+ "traffic_ratio": 4.0,
+ "traffic_rule": "innovation",
+ "traffic_seed": 4000,
+ "train_limit": 0,
+ "use_residual": 1,
+ "val_examples": 0,
+ "vectorizer_mode": "spatial_template",
+ "warmup_epochs": 0,
+ "weight_decay": 0.0001,
+ "weight_scale": 1.0,
+ "width": 16
+ },
+ "calibration_metric_space": "reciprocal_local_activity_products",
+ "counters": {
+ "apical_warmup_examples": 0,
+ "calibration_event_examples": 0,
+ "causal_scalar_observations": 0,
+ "logical_batch_loss_queries": 0,
+ "mirror_conv_examples": 0,
+ "mirror_events": 0,
+ "mirror_readout_examples": 0,
+ "neutral_projection_examples": 0,
+ "ordinary_examples": 10000000,
+ "per_example_loss_terms": 0,
+ "perturbation_events": 0,
+ "perturbation_forward_examples": 0,
+ "predictor_update_examples": 0,
+ "predictor_warmup_examples": 0,
+ "traffic_audit_examples": 0,
+ "traffic_calibration_examples": 0
+ },
+ "diagnostics": {
+ "early_third_mean": 0.9997032682100931,
+ "feedback_forward_cosine": [
+ 0.999832272529602,
+ 0.999843955039978,
+ 0.9998561143875122,
+ 0.9998370409011841,
+ 0.9998566508293152,
+ 0.9997917413711548,
+ 0.9997992515563965,
+ 0.9998595118522644,
+ 0.9998422861099243,
+ 0.9998131394386292,
+ 0.9998656511306763,
+ 0.9998103380203247,
+ 0.9998369216918945,
+ 0.999903678894043,
+ 0.9998910427093506,
+ 0.9998737573623657,
+ 0.9998822808265686,
+ 0.9997251033782959,
+ 0.9999867081642151
+ ],
+ "feedback_forward_norm_ratio": [
+ 0.9992798566818237,
+ 1.0000379085540771,
+ 0.9992694854736328,
+ 0.9998361468315125,
+ 0.9999496936798096,
+ 1.0009088516235352,
+ 0.9992625713348389,
+ 1.0002714395523071,
+ 0.9997771978378296,
+ 1.0006681680679321,
+ 0.9999299049377441,
+ 1.0009574890136719,
+ 0.9996337294578552,
+ 1.0002833604812622,
+ 1.000002145767212,
+ 1.000523567199707,
+ 1.000024676322937,
+ 1.0008426904678345,
+ 1.0024840831756592
+ ],
+ "feedback_forward_relative_error": [
+ 0.018325643613934517,
+ 0.01766648329794407,
+ 0.016976479440927505,
+ 0.018047155812382698,
+ 0.0169325340539217,
+ 0.020436281338334084,
+ 0.02004491537809372,
+ 0.016763348132371902,
+ 0.017760394141077995,
+ 0.019354287534952164,
+ 0.016392115503549576,
+ 0.0195110272616148,
+ 0.01805775612592697,
+ 0.013888871297240257,
+ 0.014755462296307087,
+ 0.015905823558568954,
+ 0.015339392237365246,
+ 0.023468511179089546,
+ 0.005726316012442112
+ ],
+ "innovation_negative_gradient_cosine": [
+ 0.9997201561927795,
+ 0.9996727705001831,
+ 0.9997096061706543,
+ 0.9996916651725769,
+ 0.9997054934501648,
+ 0.9997199177742004,
+ 0.9997192025184631,
+ 0.9997351169586182,
+ 0.9997472167015076,
+ 0.9997079372406006,
+ 0.999765157699585,
+ 0.9997378587722778,
+ 0.999783456325531,
+ 0.9998407363891602,
+ 0.9998600482940674,
+ 0.9998862743377686,
+ 0.999923825263977,
+ 0.9999722242355347,
+ 0.9151017069816589
+ ],
+ "mean_feedback_forward_cosine": 0.999847760325984,
+ "mean_feedback_forward_relative_error": 0.017123831485054995,
+ "normalization_state": "training_batch_stats_without_running_update",
+ "raw_negative_gradient_cosine": [
+ 0.9997201561927795,
+ 0.9996727705001831,
+ 0.9997096061706543,
+ 0.9996916651725769,
+ 0.9997054934501648,
+ 0.9997199177742004,
+ 0.9997192025184631,
+ 0.9997351169586182,
+ 0.9997472167015076,
+ 0.9997079372406006,
+ 0.999765157699585,
+ 0.9997378587722778,
+ 0.999783456325531,
+ 0.9998407363891602,
+ 0.9998600482940674,
+ 0.9998862743377686,
+ 0.999923825263977,
+ 0.9999722242355347,
+ 0.9151017069816589
+ ],
+ "teaching_negative_gradient_cosine": [
+ 0.9997201561927795,
+ 0.9996727705001831,
+ 0.9997096061706543,
+ 0.9996916651725769,
+ 0.9997054934501648,
+ 0.9997199177742004,
+ 0.9997192025184631,
+ 0.9997351169586182,
+ 0.9997472167015076,
+ 0.9997079372406006,
+ 0.999765157699585,
+ 0.9997378587722778,
+ 0.999783456325531,
+ 0.9998407363891602,
+ 0.9998600482940674,
+ 0.9998862743377686,
+ 0.999923825263977,
+ 0.9999722242355347,
+ 0.9151017069816589
+ ],
+ "wall_s": 0.10862064361572266
+ },
+ "epochs": [
+ {
+ "epoch": 1,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.3179482221603394,
+ "mean_feedback_forward_cosine": 0.3081172547842327,
+ "mean_feedback_forward_relative_error": 1.1762166462446515,
+ "min_feedback_forward_cosine": 0.1337927281856537
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 391,
+ "train_examples": 50000,
+ "train_loss": 1.8016380353927612
+ },
+ {
+ "epoch": 2,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.2607768774032593,
+ "mean_feedback_forward_cosine": 0.4315098618206225,
+ "mean_feedback_forward_relative_error": 1.065689112010755,
+ "min_feedback_forward_cosine": 0.2190462350845337
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 782,
+ "train_examples": 50000,
+ "train_loss": 1.3093270977020264
+ },
+ {
+ "epoch": 3,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.2074636220932007,
+ "mean_feedback_forward_cosine": 0.5265687202152453,
+ "mean_feedback_forward_relative_error": 0.9734838730410525,
+ "min_feedback_forward_cosine": 0.2893797755241394
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 1173,
+ "train_examples": 50000,
+ "train_loss": 1.0393185049438476
+ },
+ {
+ "epoch": 4,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.1569758653640747,
+ "mean_feedback_forward_cosine": 0.5972496537785781,
+ "mean_feedback_forward_relative_error": 0.8983439520785683,
+ "min_feedback_forward_cosine": 0.35155826807022095
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 1564,
+ "train_examples": 50000,
+ "train_loss": 0.8882581832504273
+ },
+ {
+ "epoch": 5,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.1111187934875488,
+ "mean_feedback_forward_cosine": 0.6488398533118399,
+ "mean_feedback_forward_relative_error": 0.8391168682198775,
+ "min_feedback_forward_cosine": 0.4040271043777466
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 1955,
+ "train_examples": 50000,
+ "train_loss": 0.7669004427909851
+ },
+ {
+ "epoch": 6,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.0666183233261108,
+ "mean_feedback_forward_cosine": 0.6898681333190516,
+ "mean_feedback_forward_relative_error": 0.7887082288139745,
+ "min_feedback_forward_cosine": 0.45204389095306396
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 2346,
+ "train_examples": 50000,
+ "train_loss": 0.6915190246772767
+ },
+ {
+ "epoch": 7,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.025307536125183,
+ "mean_feedback_forward_cosine": 0.723220386003193,
+ "mean_feedback_forward_relative_error": 0.7452961316234187,
+ "min_feedback_forward_cosine": 0.49451273679733276
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 2737,
+ "train_examples": 50000,
+ "train_loss": 0.6311031921577454
+ },
+ {
+ "epoch": 8,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.9844605326652527,
+ "mean_feedback_forward_cosine": 0.7519701907509252,
+ "mean_feedback_forward_relative_error": 0.7057159715577176,
+ "min_feedback_forward_cosine": 0.5342384576797485
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 3128,
+ "train_examples": 50000,
+ "train_loss": 0.5936740324783325
+ },
+ {
+ "epoch": 9,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.9465036988258362,
+ "mean_feedback_forward_cosine": 0.7761631419784144,
+ "mean_feedback_forward_relative_error": 0.6702790040718881,
+ "min_feedback_forward_cosine": 0.5700541734695435
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 3519,
+ "train_examples": 50000,
+ "train_loss": 0.5488329746341706
+ },
+ {
+ "epoch": 10,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.9100264310836792,
+ "mean_feedback_forward_cosine": 0.7974777974580464,
+ "mean_feedback_forward_relative_error": 0.6373831598382247,
+ "min_feedback_forward_cosine": 0.6029166579246521
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 3910,
+ "train_examples": 50000,
+ "train_loss": 0.5272661437320709
+ },
+ {
+ "epoch": 11,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.876276969909668,
+ "mean_feedback_forward_cosine": 0.8160552445210909,
+ "mean_feedback_forward_relative_error": 0.6071965616000327,
+ "min_feedback_forward_cosine": 0.6323396563529968
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 4301,
+ "train_examples": 50000,
+ "train_loss": 0.5006304055786133
+ },
+ {
+ "epoch": 12,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.8438994288444519,
+ "mean_feedback_forward_cosine": 0.8323422984073037,
+ "mean_feedback_forward_relative_error": 0.5792661622950905,
+ "min_feedback_forward_cosine": 0.6589292883872986
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 4692,
+ "train_examples": 50000,
+ "train_loss": 0.47894832363128664
+ },
+ {
+ "epoch": 13,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.812139630317688,
+ "mean_feedback_forward_cosine": 0.8468305782267922,
+ "mean_feedback_forward_relative_error": 0.5535615776714525,
+ "min_feedback_forward_cosine": 0.6840105652809143
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 5083,
+ "train_examples": 50000,
+ "train_loss": 0.4601474510383606
+ },
+ {
+ "epoch": 14,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.781139075756073,
+ "mean_feedback_forward_cosine": 0.8600040799693057,
+ "mean_feedback_forward_relative_error": 0.528963029384613,
+ "min_feedback_forward_cosine": 0.7073977589607239
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 5474,
+ "train_examples": 50000,
+ "train_loss": 0.4463700270462036
+ },
+ {
+ "epoch": 15,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.7504760026931763,
+ "mean_feedback_forward_cosine": 0.8718816387025934,
+ "mean_feedback_forward_relative_error": 0.5058058539503499,
+ "min_feedback_forward_cosine": 0.7299046516418457
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 5865,
+ "train_examples": 50000,
+ "train_loss": 0.42734941328048703
+ },
+ {
+ "epoch": 16,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.7226547598838806,
+ "mean_feedback_forward_cosine": 0.8822175013391595,
+ "mean_feedback_forward_relative_error": 0.48472224960201665,
+ "min_feedback_forward_cosine": 0.7496227025985718
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 6256,
+ "train_examples": 50000,
+ "train_loss": 0.41522703154563906
+ },
+ {
+ "epoch": 17,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.6952037811279297,
+ "mean_feedback_forward_cosine": 0.891679330875999,
+ "mean_feedback_forward_relative_error": 0.4644747729364194,
+ "min_feedback_forward_cosine": 0.7680860757827759
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 6647,
+ "train_examples": 50000,
+ "train_loss": 0.39725936639785764
+ },
+ {
+ "epoch": 18,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.6674655675888062,
+ "mean_feedback_forward_cosine": 0.9004808381984108,
+ "mean_feedback_forward_relative_error": 0.444945722034103,
+ "min_feedback_forward_cosine": 0.7860763072967529
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 7038,
+ "train_examples": 50000,
+ "train_loss": 0.39110040979385374
+ },
+ {
+ "epoch": 19,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.6415656208992004,
+ "mean_feedback_forward_cosine": 0.9084068787725348,
+ "mean_feedback_forward_relative_error": 0.42661243128149134,
+ "min_feedback_forward_cosine": 0.8021392822265625
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 7429,
+ "train_examples": 50000,
+ "train_loss": 0.38170015142440794
+ },
+ {
+ "epoch": 20,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.6167904138565063,
+ "mean_feedback_forward_cosine": 0.9155862237277784,
+ "mean_feedback_forward_relative_error": 0.409354440475765,
+ "min_feedback_forward_cosine": 0.8170633316040039
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 7820,
+ "train_examples": 50000,
+ "train_loss": 0.3708456579589844
+ },
+ {
+ "epoch": 21,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.5929568409919739,
+ "mean_feedback_forward_cosine": 0.9222921572233501,
+ "mean_feedback_forward_relative_error": 0.3925411379651019,
+ "min_feedback_forward_cosine": 0.8307660222053528
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 8211,
+ "train_examples": 50000,
+ "train_loss": 0.36358001874923707
+ },
+ {
+ "epoch": 22,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.5700664520263672,
+ "mean_feedback_forward_cosine": 0.928353770783073,
+ "mean_feedback_forward_relative_error": 0.37673597743636683,
+ "min_feedback_forward_cosine": 0.8435777425765991
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 8602,
+ "train_examples": 50000,
+ "train_loss": 0.35517278646469114
+ },
+ {
+ "epoch": 23,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.5471041798591614,
+ "mean_feedback_forward_cosine": 0.9339371166731182,
+ "mean_feedback_forward_relative_error": 0.3616281339996739,
+ "min_feedback_forward_cosine": 0.8557224273681641
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 8993,
+ "train_examples": 50000,
+ "train_loss": 0.34679950008392335
+ },
+ {
+ "epoch": 24,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.524709165096283,
+ "mean_feedback_forward_cosine": 0.9391759194825825,
+ "mean_feedback_forward_relative_error": 0.34673791496377243,
+ "min_feedback_forward_cosine": 0.8671507835388184
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 9384,
+ "train_examples": 50000,
+ "train_loss": 0.3409220140075684
+ },
+ {
+ "epoch": 25,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.5032624006271362,
+ "mean_feedback_forward_cosine": 0.943957479376542,
+ "mean_feedback_forward_relative_error": 0.3327302407277258,
+ "min_feedback_forward_cosine": 0.8777748346328735
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 9775,
+ "train_examples": 50000,
+ "train_loss": 0.3322959855651855
+ },
+ {
+ "epoch": 26,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.48299121856689453,
+ "mean_feedback_forward_cosine": 0.948312615093432,
+ "mean_feedback_forward_relative_error": 0.3194552958011627,
+ "min_feedback_forward_cosine": 0.8872740268707275
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 10166,
+ "train_examples": 50000,
+ "train_loss": 0.3208624546146393
+ },
+ {
+ "epoch": 27,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.4639672636985779,
+ "mean_feedback_forward_cosine": 0.9522071574863634,
+ "mean_feedback_forward_relative_error": 0.3070459263889413,
+ "min_feedback_forward_cosine": 0.8958839178085327
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 10557,
+ "train_examples": 50000,
+ "train_loss": 0.31833461904525756
+ },
+ {
+ "epoch": 28,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.44569382071495056,
+ "mean_feedback_forward_cosine": 0.9557621133954901,
+ "mean_feedback_forward_relative_error": 0.29537351900025416,
+ "min_feedback_forward_cosine": 0.9037723541259766
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 10948,
+ "train_examples": 50000,
+ "train_loss": 0.30910220544815065
+ },
+ {
+ "epoch": 29,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.42713597416877747,
+ "mean_feedback_forward_cosine": 0.959212811369645,
+ "mean_feedback_forward_relative_error": 0.28344978941114324,
+ "min_feedback_forward_cosine": 0.9115415811538696
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 11339,
+ "train_examples": 50000,
+ "train_loss": 0.31091059873580934
+ },
+ {
+ "epoch": 30,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.40970999002456665,
+ "mean_feedback_forward_cosine": 0.9623668319300601,
+ "mean_feedback_forward_relative_error": 0.2721605642061484,
+ "min_feedback_forward_cosine": 0.9185423851013184
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 11730,
+ "train_examples": 50000,
+ "train_loss": 0.29945821219444274
+ },
+ {
+ "epoch": 31,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.3930017650127411,
+ "mean_feedback_forward_cosine": 0.9652545138409263,
+ "mean_feedback_forward_relative_error": 0.2614814062651835,
+ "min_feedback_forward_cosine": 0.9249870181083679
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 12121,
+ "train_examples": 50000,
+ "train_loss": 0.3012135697364807
+ },
+ {
+ "epoch": 32,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.37707576155662537,
+ "mean_feedback_forward_cosine": 0.9678763872698734,
+ "mean_feedback_forward_relative_error": 0.2513225215830301,
+ "min_feedback_forward_cosine": 0.9308466911315918
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 12512,
+ "train_examples": 50000,
+ "train_loss": 0.2938556724643707
+ },
+ {
+ "epoch": 33,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.36195608973503113,
+ "mean_feedback_forward_cosine": 0.9702881135438618,
+ "mean_feedback_forward_relative_error": 0.24166028711356616,
+ "min_feedback_forward_cosine": 0.9362145662307739
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 12903,
+ "train_examples": 50000,
+ "train_loss": 0.28622951061725616
+ },
+ {
+ "epoch": 34,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.34745699167251587,
+ "mean_feedback_forward_cosine": 0.9725221520976016,
+ "mean_feedback_forward_relative_error": 0.23229151258343145,
+ "min_feedback_forward_cosine": 0.9411567449569702
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 13294,
+ "train_examples": 50000,
+ "train_loss": 0.28590569502830504
+ },
+ {
+ "epoch": 35,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.3339449465274811,
+ "mean_feedback_forward_cosine": 0.9745953459488718,
+ "mean_feedback_forward_relative_error": 0.22325121493715988,
+ "min_feedback_forward_cosine": 0.9455727338790894
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 13685,
+ "train_examples": 50000,
+ "train_loss": 0.27529972619056703
+ },
+ {
+ "epoch": 36,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.32066166400909424,
+ "mean_feedback_forward_cosine": 0.9765140759317499,
+ "mean_feedback_forward_relative_error": 0.21460931042307302,
+ "min_feedback_forward_cosine": 0.9497660398483276
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 14076,
+ "train_examples": 50000,
+ "train_loss": 0.27331757281303404
+ },
+ {
+ "epoch": 37,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.3072468340396881,
+ "mean_feedback_forward_cosine": 0.9783117551552621,
+ "mean_feedback_forward_relative_error": 0.2061853212745566,
+ "min_feedback_forward_cosine": 0.953851044178009
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 14467,
+ "train_examples": 50000,
+ "train_loss": 0.2763000616645813
+ },
+ {
+ "epoch": 38,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.2946045696735382,
+ "mean_feedback_forward_cosine": 0.9799527274934869,
+ "mean_feedback_forward_relative_error": 0.1981760110510023,
+ "min_feedback_forward_cosine": 0.9575166702270508
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 14858,
+ "train_examples": 50000,
+ "train_loss": 0.2736293261241913
+ },
+ {
+ "epoch": 39,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.28286978602409363,
+ "mean_feedback_forward_cosine": 0.9814589964716058,
+ "mean_feedback_forward_relative_error": 0.19053916123352552,
+ "min_feedback_forward_cosine": 0.9608039855957031
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 15249,
+ "train_examples": 50000,
+ "train_loss": 0.26910734585762025
+ },
+ {
+ "epoch": 40,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.271408349275589,
+ "mean_feedback_forward_cosine": 0.9828565183438753,
+ "mean_feedback_forward_relative_error": 0.18318781374316467,
+ "min_feedback_forward_cosine": 0.9638869762420654
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 15640,
+ "train_examples": 50000,
+ "train_loss": 0.26631804656505587
+ },
+ {
+ "epoch": 41,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.26066309213638306,
+ "mean_feedback_forward_cosine": 0.9841489791870117,
+ "mean_feedback_forward_relative_error": 0.1760815589835769,
+ "min_feedback_forward_cosine": 0.9666633605957031
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 16031,
+ "train_examples": 50000,
+ "train_loss": 0.2643160597515106
+ },
+ {
+ "epoch": 42,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.25069287419319153,
+ "mean_feedback_forward_cosine": 0.9853054222307707,
+ "mean_feedback_forward_relative_error": 0.1694794583477472,
+ "min_feedback_forward_cosine": 0.9691227078437805
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 16422,
+ "train_examples": 50000,
+ "train_loss": 0.2576930554866791
+ },
+ {
+ "epoch": 43,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.24042922258377075,
+ "mean_feedback_forward_cosine": 0.9864118130583512,
+ "mean_feedback_forward_relative_error": 0.16292876279667803,
+ "min_feedback_forward_cosine": 0.9715813994407654
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 16813,
+ "train_examples": 50000,
+ "train_loss": 0.25705379019737246
+ },
+ {
+ "epoch": 44,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.2306688129901886,
+ "mean_feedback_forward_cosine": 0.9874522874229833,
+ "mean_feedback_forward_relative_error": 0.15654690563678741,
+ "min_feedback_forward_cosine": 0.9738315939903259
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 17204,
+ "train_examples": 50000,
+ "train_loss": 0.2570527005672455
+ },
+ {
+ "epoch": 45,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.22168520092964172,
+ "mean_feedback_forward_cosine": 0.9883866937536943,
+ "mean_feedback_forward_relative_error": 0.15057311952114105,
+ "min_feedback_forward_cosine": 0.9758184552192688
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 17595,
+ "train_examples": 50000,
+ "train_loss": 0.2519144667339325
+ },
+ {
+ "epoch": 46,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.21260973811149597,
+ "mean_feedback_forward_cosine": 0.9892514065692299,
+ "mean_feedback_forward_relative_error": 0.14482591497270683,
+ "min_feedback_forward_cosine": 0.9777439832687378
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 17986,
+ "train_examples": 50000,
+ "train_loss": 0.2517258186197281
+ },
+ {
+ "epoch": 47,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.2041763961315155,
+ "mean_feedback_forward_cosine": 0.9900465576272262,
+ "mean_feedback_forward_relative_error": 0.1393495253042171,
+ "min_feedback_forward_cosine": 0.9794571399688721
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 18377,
+ "train_examples": 50000,
+ "train_loss": 0.24572330734252928
+ },
+ {
+ "epoch": 48,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.19572725892066956,
+ "mean_feedback_forward_cosine": 0.9907976671269065,
+ "mean_feedback_forward_relative_error": 0.13397606501453801,
+ "min_feedback_forward_cosine": 0.9811136722564697
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 18768,
+ "train_examples": 50000,
+ "train_loss": 0.25035988627910616
+ },
+ {
+ "epoch": 49,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.1879836618900299,
+ "mean_feedback_forward_cosine": 0.9914799991406893,
+ "mean_feedback_forward_relative_error": 0.128885799136601,
+ "min_feedback_forward_cosine": 0.9825654029846191
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 19159,
+ "train_examples": 50000,
+ "train_loss": 0.24520568934440612
+ },
+ {
+ "epoch": 50,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.18033981323242188,
+ "mean_feedback_forward_cosine": 0.9921371623089439,
+ "mean_feedback_forward_relative_error": 0.12380391418149597,
+ "min_feedback_forward_cosine": 0.983942985534668
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 19550,
+ "train_examples": 50000,
+ "train_loss": 0.24459320890426636
+ },
+ {
+ "epoch": 51,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.17296352982521057,
+ "mean_feedback_forward_cosine": 0.9927325687910381,
+ "mean_feedback_forward_relative_error": 0.1190079266303464,
+ "min_feedback_forward_cosine": 0.9852181077003479
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 19941,
+ "train_examples": 50000,
+ "train_loss": 0.24023687831401824
+ },
+ {
+ "epoch": 52,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.16581204533576965,
+ "mean_feedback_forward_cosine": 0.9932821926317716,
+ "mean_feedback_forward_relative_error": 0.11440283275748554,
+ "min_feedback_forward_cosine": 0.9864091873168945
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 20332,
+ "train_examples": 50000,
+ "train_loss": 0.23881971601486207
+ },
+ {
+ "epoch": 53,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.1592707484960556,
+ "mean_feedback_forward_cosine": 0.9937764531687686,
+ "mean_feedback_forward_relative_error": 0.11010334876022841,
+ "min_feedback_forward_cosine": 0.9874594211578369
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 20723,
+ "train_examples": 50000,
+ "train_loss": 0.23764040531158448
+ },
+ {
+ "epoch": 54,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.15281471610069275,
+ "mean_feedback_forward_cosine": 0.9942525938937539,
+ "mean_feedback_forward_relative_error": 0.10579933677064746,
+ "min_feedback_forward_cosine": 0.9884459376335144
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 21114,
+ "train_examples": 50000,
+ "train_loss": 0.23659150759220124
+ },
+ {
+ "epoch": 55,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.14686770737171173,
+ "mean_feedback_forward_cosine": 0.9946752190589905,
+ "mean_feedback_forward_relative_error": 0.10181317713699843,
+ "min_feedback_forward_cosine": 0.9893269538879395
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 21505,
+ "train_examples": 50000,
+ "train_loss": 0.23042542052745818
+ },
+ {
+ "epoch": 56,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.14074470102787018,
+ "mean_feedback_forward_cosine": 0.9950867514861258,
+ "mean_feedback_forward_relative_error": 0.09778856878217898,
+ "min_feedback_forward_cosine": 0.9901920557022095
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 21896,
+ "train_examples": 50000,
+ "train_loss": 0.23666077642440797
+ },
+ {
+ "epoch": 57,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.13506698608398438,
+ "mean_feedback_forward_cosine": 0.9954530439878765,
+ "mean_feedback_forward_relative_error": 0.09406332455967602,
+ "min_feedback_forward_cosine": 0.9909622669219971
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 22287,
+ "train_examples": 50000,
+ "train_loss": 0.2333391785812378
+ },
+ {
+ "epoch": 58,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.12974236905574799,
+ "mean_feedback_forward_cosine": 0.995790277656756,
+ "mean_feedback_forward_relative_error": 0.09050365360943895,
+ "min_feedback_forward_cosine": 0.9916596412658691
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 22678,
+ "train_examples": 50000,
+ "train_loss": 0.2275290870332718
+ },
+ {
+ "epoch": 59,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.1242767944931984,
+ "mean_feedback_forward_cosine": 0.9961188121845848,
+ "mean_feedback_forward_relative_error": 0.08691137617355899,
+ "min_feedback_forward_cosine": 0.9923429489135742
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 23069,
+ "train_examples": 50000,
+ "train_loss": 0.23219734165668487
+ },
+ {
+ "epoch": 60,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.11934433877468109,
+ "mean_feedback_forward_cosine": 0.9964093848278648,
+ "mean_feedback_forward_relative_error": 0.08358648321346233,
+ "min_feedback_forward_cosine": 0.9929372668266296
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 23460,
+ "train_examples": 50000,
+ "train_loss": 0.22319554695129395
+ },
+ {
+ "epoch": 61,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.11451341211795807,
+ "mean_feedback_forward_cosine": 0.9966863927088285,
+ "mean_feedback_forward_relative_error": 0.08030059894448832,
+ "min_feedback_forward_cosine": 0.993494987487793
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 23851,
+ "train_examples": 50000,
+ "train_loss": 0.22576308106422424
+ },
+ {
+ "epoch": 62,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.10992468148469925,
+ "mean_feedback_forward_cosine": 0.9969351197543898,
+ "mean_feedback_forward_relative_error": 0.07722324544661924,
+ "min_feedback_forward_cosine": 0.9940042495727539
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 24242,
+ "train_examples": 50000,
+ "train_loss": 0.2229993958377838
+ },
+ {
+ "epoch": 63,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.1056821271777153,
+ "mean_feedback_forward_cosine": 0.9971621569834257,
+ "mean_feedback_forward_relative_error": 0.0742951021774819,
+ "min_feedback_forward_cosine": 0.9944562911987305
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 24633,
+ "train_examples": 50000,
+ "train_loss": 0.21717544426918028
+ },
+ {
+ "epoch": 64,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.10151567310094833,
+ "mean_feedback_forward_cosine": 0.9973798017752799,
+ "mean_feedback_forward_relative_error": 0.07139499642347034,
+ "min_feedback_forward_cosine": 0.9948840141296387
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 25024,
+ "train_examples": 50000,
+ "train_loss": 0.22616530433654786
+ },
+ {
+ "epoch": 65,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0975341796875,
+ "mean_feedback_forward_cosine": 0.9975817297634325,
+ "mean_feedback_forward_relative_error": 0.06859047063871433,
+ "min_feedback_forward_cosine": 0.9952762126922607
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 25415,
+ "train_examples": 50000,
+ "train_loss": 0.2245164638710022
+ },
+ {
+ "epoch": 66,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.09380342066287994,
+ "mean_feedback_forward_cosine": 0.9977607381971259,
+ "mean_feedback_forward_relative_error": 0.06598612332814618,
+ "min_feedback_forward_cosine": 0.9956289529800415
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 25806,
+ "train_examples": 50000,
+ "train_loss": 0.21714474848747253
+ },
+ {
+ "epoch": 67,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.09001902490854263,
+ "mean_feedback_forward_cosine": 0.9979298459856134,
+ "mean_feedback_forward_relative_error": 0.06344172927109819,
+ "min_feedback_forward_cosine": 0.9959728717803955
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 26197,
+ "train_examples": 50000,
+ "train_loss": 0.22183403856277467
+ },
+ {
+ "epoch": 68,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.08653908967971802,
+ "mean_feedback_forward_cosine": 0.9980833624538622,
+ "mean_feedback_forward_relative_error": 0.061045237767853235,
+ "min_feedback_forward_cosine": 0.9962779879570007
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 26588,
+ "train_examples": 50000,
+ "train_loss": 0.21689167611122132
+ },
+ {
+ "epoch": 69,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.08311554044485092,
+ "mean_feedback_forward_cosine": 0.9982279633220873,
+ "mean_feedback_forward_relative_error": 0.058698259392067006,
+ "min_feedback_forward_cosine": 0.9965658783912659
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 26979,
+ "train_examples": 50000,
+ "train_loss": 0.21927908441066743
+ },
+ {
+ "epoch": 70,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.07982558012008667,
+ "mean_feedback_forward_cosine": 0.998361443218432,
+ "mean_feedback_forward_relative_error": 0.056440566617407296,
+ "min_feedback_forward_cosine": 0.9968308806419373
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 27370,
+ "train_examples": 50000,
+ "train_loss": 0.21678533363342284
+ },
+ {
+ "epoch": 71,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.07658115774393082,
+ "mean_feedback_forward_cosine": 0.9984848373814633,
+ "mean_feedback_forward_relative_error": 0.05427446726121401,
+ "min_feedback_forward_cosine": 0.9970823526382446
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 27761,
+ "train_examples": 50000,
+ "train_loss": 0.21589837124824524
+ },
+ {
+ "epoch": 72,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.07362756133079529,
+ "mean_feedback_forward_cosine": 0.998594770306035,
+ "mean_feedback_forward_relative_error": 0.05225870758295059,
+ "min_feedback_forward_cosine": 0.9973031282424927
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 28152,
+ "train_examples": 50000,
+ "train_loss": 0.21279439496994018
+ },
+ {
+ "epoch": 73,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.07068312913179398,
+ "mean_feedback_forward_cosine": 0.9987004618895682,
+ "mean_feedback_forward_relative_error": 0.050256861177714246,
+ "min_feedback_forward_cosine": 0.9975138306617737
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 28543,
+ "train_examples": 50000,
+ "train_loss": 0.21149720972061156
+ },
+ {
+ "epoch": 74,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.06784600019454956,
+ "mean_feedback_forward_cosine": 0.9987988346501401,
+ "mean_feedback_forward_relative_error": 0.048314295807167104,
+ "min_feedback_forward_cosine": 0.9977090954780579
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 28934,
+ "train_examples": 50000,
+ "train_loss": 0.21316380685806274
+ },
+ {
+ "epoch": 75,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.06512412428855896,
+ "mean_feedback_forward_cosine": 0.9988879153603002,
+ "mean_feedback_forward_relative_error": 0.04648596460097715,
+ "min_feedback_forward_cosine": 0.997888445854187
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 29325,
+ "train_examples": 50000,
+ "train_loss": 0.21025281191825868
+ },
+ {
+ "epoch": 76,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0625939890742302,
+ "mean_feedback_forward_cosine": 0.998970301527726,
+ "mean_feedback_forward_relative_error": 0.04472720456358634,
+ "min_feedback_forward_cosine": 0.9980493783950806
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 29716,
+ "train_examples": 50000,
+ "train_loss": 0.20640591926574706
+ },
+ {
+ "epoch": 77,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.05987943336367607,
+ "mean_feedback_forward_cosine": 0.9990504635007758,
+ "mean_feedback_forward_relative_error": 0.04295391452155615,
+ "min_feedback_forward_cosine": 0.998214602470398
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 30107,
+ "train_examples": 50000,
+ "train_loss": 0.2156642873096466
+ },
+ {
+ "epoch": 78,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.05744881182909012,
+ "mean_feedback_forward_cosine": 0.9991243701232108,
+ "mean_feedback_forward_relative_error": 0.04125269610238703,
+ "min_feedback_forward_cosine": 0.9983565807342529
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 30498,
+ "train_examples": 50000,
+ "train_loss": 0.21317154891014098
+ },
+ {
+ "epoch": 79,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.05520369112491608,
+ "mean_feedback_forward_cosine": 0.9991884168825651,
+ "mean_feedback_forward_relative_error": 0.0397133147834163,
+ "min_feedback_forward_cosine": 0.998482346534729
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 30889,
+ "train_examples": 50000,
+ "train_loss": 0.2100965942764282
+ },
+ {
+ "epoch": 80,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0530511774122715,
+ "mean_feedback_forward_cosine": 0.9992481877929286,
+ "mean_feedback_forward_relative_error": 0.03821724751277974,
+ "min_feedback_forward_cosine": 0.9985981583595276
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 31280,
+ "train_examples": 50000,
+ "train_loss": 0.20471788811683655
+ },
+ {
+ "epoch": 81,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.05102203041315079,
+ "mean_feedback_forward_cosine": 0.9993047651491667,
+ "mean_feedback_forward_relative_error": 0.03675123823708609,
+ "min_feedback_forward_cosine": 0.9987033605575562
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 31671,
+ "train_examples": 50000,
+ "train_loss": 0.2067176330423355
+ },
+ {
+ "epoch": 82,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.04897250980138779,
+ "mean_feedback_forward_cosine": 0.9993574305584556,
+ "mean_feedback_forward_relative_error": 0.035332459760339635,
+ "min_feedback_forward_cosine": 0.9988054037094116
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 32062,
+ "train_examples": 50000,
+ "train_loss": 0.2053587057876587
+ },
+ {
+ "epoch": 83,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.04702990502119064,
+ "mean_feedback_forward_cosine": 0.9994063565605565,
+ "mean_feedback_forward_relative_error": 0.03396036010235548,
+ "min_feedback_forward_cosine": 0.9988979697227478
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 32453,
+ "train_examples": 50000,
+ "train_loss": 0.20637111597061158
+ },
+ {
+ "epoch": 84,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.04517826810479164,
+ "mean_feedback_forward_cosine": 0.9994505361506814,
+ "mean_feedback_forward_relative_error": 0.03267086346290613,
+ "min_feedback_forward_cosine": 0.9989829063415527
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 32844,
+ "train_examples": 50000,
+ "train_loss": 0.2027370511198044
+ },
+ {
+ "epoch": 85,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.04336803779006004,
+ "mean_feedback_forward_cosine": 0.9994919864754928,
+ "mean_feedback_forward_relative_error": 0.031413392604965916,
+ "min_feedback_forward_cosine": 0.9990625381469727
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 33235,
+ "train_examples": 50000,
+ "train_loss": 0.20505338558197023
+ },
+ {
+ "epoch": 86,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.04163755476474762,
+ "mean_feedback_forward_cosine": 0.9995299609083879,
+ "mean_feedback_forward_relative_error": 0.030214958530115455,
+ "min_feedback_forward_cosine": 0.9991358518600464
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 33626,
+ "train_examples": 50000,
+ "train_loss": 0.20693019121170045
+ },
+ {
+ "epoch": 87,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03998049721121788,
+ "mean_feedback_forward_cosine": 0.9995655135104531,
+ "mean_feedback_forward_relative_error": 0.029049062758292023,
+ "min_feedback_forward_cosine": 0.9992032647132874
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 34017,
+ "train_examples": 50000,
+ "train_loss": 0.20767693899154663
+ },
+ {
+ "epoch": 88,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03840848058462143,
+ "mean_feedback_forward_cosine": 0.9995980231385482,
+ "mean_feedback_forward_relative_error": 0.027941891041241194,
+ "min_feedback_forward_cosine": 0.9992645978927612
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 34408,
+ "train_examples": 50000,
+ "train_loss": 0.20105957365989685
+ },
+ {
+ "epoch": 89,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03683806210756302,
+ "mean_feedback_forward_cosine": 0.9996288701107627,
+ "mean_feedback_forward_relative_error": 0.02684645134171373,
+ "min_feedback_forward_cosine": 0.9993234872817993
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 34799,
+ "train_examples": 50000,
+ "train_loss": 0.20099872408390046
+ },
+ {
+ "epoch": 90,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03540956601500511,
+ "mean_feedback_forward_cosine": 0.9996562129572818,
+ "mean_feedback_forward_relative_error": 0.02584206317796519,
+ "min_feedback_forward_cosine": 0.9993747472763062
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 35190,
+ "train_examples": 50000,
+ "train_loss": 0.20426013642787932
+ },
+ {
+ "epoch": 91,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03400075063109398,
+ "mean_feedback_forward_cosine": 0.9996820311797293,
+ "mean_feedback_forward_relative_error": 0.02485535259505636,
+ "min_feedback_forward_cosine": 0.9994235634803772
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 35581,
+ "train_examples": 50000,
+ "train_loss": 0.20525977241992951
+ },
+ {
+ "epoch": 92,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03268192708492279,
+ "mean_feedback_forward_cosine": 0.9997056816753588,
+ "mean_feedback_forward_relative_error": 0.023911231148399804,
+ "min_feedback_forward_cosine": 0.9994674324989319
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 35972,
+ "train_examples": 50000,
+ "train_loss": 0.1998370484352112
+ },
+ {
+ "epoch": 93,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03140087425708771,
+ "mean_feedback_forward_cosine": 0.9997279236191198,
+ "mean_feedback_forward_relative_error": 0.022987395770063524,
+ "min_feedback_forward_cosine": 0.999508261680603
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 36363,
+ "train_examples": 50000,
+ "train_loss": 0.19679815489768981
+ },
+ {
+ "epoch": 94,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.030177893117070198,
+ "mean_feedback_forward_cosine": 0.999748386834797,
+ "mean_feedback_forward_relative_error": 0.02210633040062691,
+ "min_feedback_forward_cosine": 0.9995459318161011
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 36754,
+ "train_examples": 50000,
+ "train_loss": 0.2004633063650131
+ },
+ {
+ "epoch": 95,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.029030324891209602,
+ "mean_feedback_forward_cosine": 0.9997664062600387,
+ "mean_feedback_forward_relative_error": 0.02129706652148774,
+ "min_feedback_forward_cosine": 0.9995797872543335
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 37145,
+ "train_examples": 50000,
+ "train_loss": 0.19316452387809754
+ },
+ {
+ "epoch": 96,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02786830998957157,
+ "mean_feedback_forward_cosine": 0.9997845417574832,
+ "mean_feedback_forward_relative_error": 0.02045521355773273,
+ "min_feedback_forward_cosine": 0.99961256980896
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 37536,
+ "train_examples": 50000,
+ "train_loss": 0.1993290764427185
+ },
+ {
+ "epoch": 97,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.026795070618391037,
+ "mean_feedback_forward_cosine": 0.999800914212277,
+ "mean_feedback_forward_relative_error": 0.01966134963655158,
+ "min_feedback_forward_cosine": 0.9996418952941895
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 37927,
+ "train_examples": 50000,
+ "train_loss": 0.19722025142669677
+ },
+ {
+ "epoch": 98,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.025728952139616013,
+ "mean_feedback_forward_cosine": 0.9998162953477157,
+ "mean_feedback_forward_relative_error": 0.018887894032032865,
+ "min_feedback_forward_cosine": 0.9996697902679443
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 38318,
+ "train_examples": 50000,
+ "train_loss": 0.20411555974960327
+ },
+ {
+ "epoch": 99,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.024722104892134666,
+ "mean_feedback_forward_cosine": 0.9998298914808976,
+ "mean_feedback_forward_relative_error": 0.018175248350752026,
+ "min_feedback_forward_cosine": 0.9996950626373291
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 38709,
+ "train_examples": 50000,
+ "train_loss": 0.19462783898353578
+ },
+ {
+ "epoch": 100,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023741193115711212,
+ "mean_feedback_forward_cosine": 0.9998429637206229,
+ "mean_feedback_forward_relative_error": 0.017462985124439,
+ "min_feedback_forward_cosine": 0.9997187852859497
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 39100,
+ "train_examples": 50000,
+ "train_loss": 0.1922397227048874
+ },
+ {
+ "epoch": 101,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023732464760541916,
+ "mean_feedback_forward_cosine": 0.9998430923411721,
+ "mean_feedback_forward_relative_error": 0.017453264298015518,
+ "min_feedback_forward_cosine": 0.9997190237045288
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 39491,
+ "train_examples": 50000,
+ "train_loss": 0.12268428046226501
+ },
+ {
+ "epoch": 102,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023727644234895706,
+ "mean_feedback_forward_cosine": 0.9998431456716437,
+ "mean_feedback_forward_relative_error": 0.017445796267374566,
+ "min_feedback_forward_cosine": 0.999718964099884
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 39882,
+ "train_examples": 50000,
+ "train_loss": 0.09654224486351014
+ },
+ {
+ "epoch": 103,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02372295968234539,
+ "mean_feedback_forward_cosine": 0.9998432366471541,
+ "mean_feedback_forward_relative_error": 0.017438747356400677,
+ "min_feedback_forward_cosine": 0.9997191429138184
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 40273,
+ "train_examples": 50000,
+ "train_loss": 0.08849230135440826
+ },
+ {
+ "epoch": 104,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023718351498246193,
+ "mean_feedback_forward_cosine": 0.9998433401710108,
+ "mean_feedback_forward_relative_error": 0.017431871622408692,
+ "min_feedback_forward_cosine": 0.9997193217277527
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 40664,
+ "train_examples": 50000,
+ "train_loss": 0.08014775415182114
+ },
+ {
+ "epoch": 105,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02371394820511341,
+ "mean_feedback_forward_cosine": 0.999843418598175,
+ "mean_feedback_forward_relative_error": 0.017425256756771552,
+ "min_feedback_forward_cosine": 0.9997194409370422
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 41055,
+ "train_examples": 50000,
+ "train_loss": 0.07335598557710647
+ },
+ {
+ "epoch": 106,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023709459230303764,
+ "mean_feedback_forward_cosine": 0.9998434938882527,
+ "mean_feedback_forward_relative_error": 0.017418845630201855,
+ "min_feedback_forward_cosine": 0.999719500541687
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 41446,
+ "train_examples": 50000,
+ "train_loss": 0.07254063300132751
+ },
+ {
+ "epoch": 107,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023704800754785538,
+ "mean_feedback_forward_cosine": 0.999843556629984,
+ "mean_feedback_forward_relative_error": 0.01741234208212087,
+ "min_feedback_forward_cosine": 0.9997196197509766
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 41837,
+ "train_examples": 50000,
+ "train_loss": 0.06800365599751472
+ },
+ {
+ "epoch": 108,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023700280115008354,
+ "mean_feedback_forward_cosine": 0.9998436601538407,
+ "mean_feedback_forward_relative_error": 0.017406211553239508,
+ "min_feedback_forward_cosine": 0.9997197389602661
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 42228,
+ "train_examples": 50000,
+ "train_loss": 0.06385894116640091
+ },
+ {
+ "epoch": 109,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023695804178714752,
+ "mean_feedback_forward_cosine": 0.9998437636776975,
+ "mean_feedback_forward_relative_error": 0.01740002825758175,
+ "min_feedback_forward_cosine": 0.9997198581695557
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 42619,
+ "train_examples": 50000,
+ "train_loss": 0.06184221161365509
+ },
+ {
+ "epoch": 110,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023691022768616676,
+ "mean_feedback_forward_cosine": 0.9998438326936019,
+ "mean_feedback_forward_relative_error": 0.01739368033840468,
+ "min_feedback_forward_cosine": 0.99972003698349
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 43010,
+ "train_examples": 50000,
+ "train_loss": 0.060633849470615384
+ },
+ {
+ "epoch": 111,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023686127737164497,
+ "mean_feedback_forward_cosine": 0.9998439236691123,
+ "mean_feedback_forward_relative_error": 0.01738741413052929,
+ "min_feedback_forward_cosine": 0.9997200965881348
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 43401,
+ "train_examples": 50000,
+ "train_loss": 0.060744912099838254
+ },
+ {
+ "epoch": 112,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023681461811065674,
+ "mean_feedback_forward_cosine": 0.9998440115075362,
+ "mean_feedback_forward_relative_error": 0.017381269092622557,
+ "min_feedback_forward_cosine": 0.9997202157974243
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 43792,
+ "train_examples": 50000,
+ "train_loss": 0.056162936646938326
+ },
+ {
+ "epoch": 113,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023676710203289986,
+ "mean_feedback_forward_cosine": 0.9998440617009213,
+ "mean_feedback_forward_relative_error": 0.01737514133320043,
+ "min_feedback_forward_cosine": 0.9997203350067139
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 44183,
+ "train_examples": 50000,
+ "train_loss": 0.054906321696043016
+ },
+ {
+ "epoch": 114,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02367215044796467,
+ "mean_feedback_forward_cosine": 0.9998441683618646,
+ "mean_feedback_forward_relative_error": 0.0173690825406658,
+ "min_feedback_forward_cosine": 0.9997203946113586
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 44574,
+ "train_examples": 50000,
+ "train_loss": 0.05158617860794067
+ },
+ {
+ "epoch": 115,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02366732992231846,
+ "mean_feedback_forward_cosine": 0.9998442373777691,
+ "mean_feedback_forward_relative_error": 0.01736301357710832,
+ "min_feedback_forward_cosine": 0.9997204542160034
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 44965,
+ "train_examples": 50000,
+ "train_loss": 0.05088043869256973
+ },
+ {
+ "epoch": 116,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023662425577640533,
+ "mean_feedback_forward_cosine": 0.9998443252161929,
+ "mean_feedback_forward_relative_error": 0.01735682993833172,
+ "min_feedback_forward_cosine": 0.9997206926345825
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 45356,
+ "train_examples": 50000,
+ "train_loss": 0.049507606768608096
+ },
+ {
+ "epoch": 117,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023657245561480522,
+ "mean_feedback_forward_cosine": 0.999844397369184,
+ "mean_feedback_forward_relative_error": 0.0173505259138581,
+ "min_feedback_forward_cosine": 0.9997208118438721
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 45747,
+ "train_examples": 50000,
+ "train_loss": 0.048889036183357236
+ },
+ {
+ "epoch": 118,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02365242876112461,
+ "mean_feedback_forward_cosine": 0.9998445259897333,
+ "mean_feedback_forward_relative_error": 0.01734436661201088,
+ "min_feedback_forward_cosine": 0.9997209310531616
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 46138,
+ "train_examples": 50000,
+ "train_loss": 0.046154447889328
+ },
+ {
+ "epoch": 119,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023647358641028404,
+ "mean_feedback_forward_cosine": 0.9998445761831183,
+ "mean_feedback_forward_relative_error": 0.017338183659471963,
+ "min_feedback_forward_cosine": 0.9997210502624512
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 46529,
+ "train_examples": 50000,
+ "train_loss": 0.045580965416431425
+ },
+ {
+ "epoch": 120,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023642217740416527,
+ "mean_feedback_forward_cosine": 0.9998446671586287,
+ "mean_feedback_forward_relative_error": 0.017331985683229408,
+ "min_feedback_forward_cosine": 0.999721109867096
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 46920,
+ "train_examples": 50000,
+ "train_loss": 0.044149961864948274
+ },
+ {
+ "epoch": 121,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023637007921934128,
+ "mean_feedback_forward_cosine": 0.9998447957791781,
+ "mean_feedback_forward_relative_error": 0.017325968261023884,
+ "min_feedback_forward_cosine": 0.9997212290763855
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 47311,
+ "train_examples": 50000,
+ "train_loss": 0.04359179922521114
+ },
+ {
+ "epoch": 122,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023631718009710312,
+ "mean_feedback_forward_cosine": 0.99984483969839,
+ "mean_feedback_forward_relative_error": 0.017319751094634597,
+ "min_feedback_forward_cosine": 0.999721348285675
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 47702,
+ "train_examples": 50000,
+ "train_loss": 0.04161858483195305
+ },
+ {
+ "epoch": 123,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023626845329999924,
+ "mean_feedback_forward_cosine": 0.9998449212626407,
+ "mean_feedback_forward_relative_error": 0.01731370088007105,
+ "min_feedback_forward_cosine": 0.9997215270996094
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 48093,
+ "train_examples": 50000,
+ "train_loss": 0.038731798566579816
+ },
+ {
+ "epoch": 124,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02362195774912834,
+ "mean_feedback_forward_cosine": 0.9998450122381511,
+ "mean_feedback_forward_relative_error": 0.017307680835457223,
+ "min_feedback_forward_cosine": 0.9997216463088989
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 48484,
+ "train_examples": 50000,
+ "train_loss": 0.038410091083049776
+ },
+ {
+ "epoch": 125,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02361702173948288,
+ "mean_feedback_forward_cosine": 0.9998450938024019,
+ "mean_feedback_forward_relative_error": 0.01730163615981215,
+ "min_feedback_forward_cosine": 0.9997217655181885
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 48875,
+ "train_examples": 50000,
+ "train_loss": 0.03799481664180756
+ },
+ {
+ "epoch": 126,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023611996322870255,
+ "mean_feedback_forward_cosine": 0.9998451847779123,
+ "mean_feedback_forward_relative_error": 0.017295557784994964,
+ "min_feedback_forward_cosine": 0.999721884727478
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 49266,
+ "train_examples": 50000,
+ "train_loss": 0.037503372143507005
+ },
+ {
+ "epoch": 127,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023606695234775543,
+ "mean_feedback_forward_cosine": 0.9998452632050765,
+ "mean_feedback_forward_relative_error": 0.017289149035748682,
+ "min_feedback_forward_cosine": 0.9997219443321228
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 49657,
+ "train_examples": 50000,
+ "train_loss": 0.03679502383947372
+ },
+ {
+ "epoch": 128,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023601356893777847,
+ "mean_feedback_forward_cosine": 0.999845382414366,
+ "mean_feedback_forward_relative_error": 0.017282778299168536,
+ "min_feedback_forward_cosine": 0.9997220039367676
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 50048,
+ "train_examples": 50000,
+ "train_loss": 0.03629144645452499
+ },
+ {
+ "epoch": 129,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02359592169523239,
+ "mean_feedback_forward_cosine": 0.9998454828011362,
+ "mean_feedback_forward_relative_error": 0.017276233111165072,
+ "min_feedback_forward_cosine": 0.9997222423553467
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 50439,
+ "train_examples": 50000,
+ "train_loss": 0.0364483181989193
+ },
+ {
+ "epoch": 130,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02359072118997574,
+ "mean_feedback_forward_cosine": 0.9998455518170407,
+ "mean_feedback_forward_relative_error": 0.017269935434390055,
+ "min_feedback_forward_cosine": 0.9997223019599915
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 50830,
+ "train_examples": 50000,
+ "train_loss": 0.034332342364788056
+ },
+ {
+ "epoch": 131,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02358531393110752,
+ "mean_feedback_forward_cosine": 0.9998456961230228,
+ "mean_feedback_forward_relative_error": 0.01726341384806131,
+ "min_feedback_forward_cosine": 0.9997224807739258
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 51221,
+ "train_examples": 50000,
+ "train_loss": 0.03428687328636646
+ },
+ {
+ "epoch": 132,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02358020283281803,
+ "mean_feedback_forward_cosine": 0.9998457682760138,
+ "mean_feedback_forward_relative_error": 0.01725717761406773,
+ "min_feedback_forward_cosine": 0.9997225999832153
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 51612,
+ "train_examples": 50000,
+ "train_loss": 0.032234325948953625
+ },
+ {
+ "epoch": 133,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023574132472276688,
+ "mean_feedback_forward_cosine": 0.9998458561144377,
+ "mean_feedback_forward_relative_error": 0.017250330775584046,
+ "min_feedback_forward_cosine": 0.9997227191925049
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 52003,
+ "train_examples": 50000,
+ "train_loss": 0.034230175194740295
+ },
+ {
+ "epoch": 134,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023568494245409966,
+ "mean_feedback_forward_cosine": 0.9998459753237272,
+ "mean_feedback_forward_relative_error": 0.017243789778532165,
+ "min_feedback_forward_cosine": 0.9997227787971497
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 52394,
+ "train_examples": 50000,
+ "train_loss": 0.03254088444650173
+ },
+ {
+ "epoch": 135,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023562785238027573,
+ "mean_feedback_forward_cosine": 0.9998460506138048,
+ "mean_feedback_forward_relative_error": 0.017237197926366014,
+ "min_feedback_forward_cosine": 0.999722957611084
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 52785,
+ "train_examples": 50000,
+ "train_loss": 0.03148084326982498
+ },
+ {
+ "epoch": 136,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023556891828775406,
+ "mean_feedback_forward_cosine": 0.9998461792343541,
+ "mean_feedback_forward_relative_error": 0.01723046524842319,
+ "min_feedback_forward_cosine": 0.9997230768203735
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 53176,
+ "train_examples": 50000,
+ "train_loss": 0.03139829716026783
+ },
+ {
+ "epoch": 137,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023550614714622498,
+ "mean_feedback_forward_cosine": 0.9998462796211243,
+ "mean_feedback_forward_relative_error": 0.01722343726769874,
+ "min_feedback_forward_cosine": 0.9997232556343079
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 53567,
+ "train_examples": 50000,
+ "train_loss": 0.03205857029914856
+ },
+ {
+ "epoch": 138,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023544590920209885,
+ "mean_feedback_forward_cosine": 0.9998463894191542,
+ "mean_feedback_forward_relative_error": 0.017216602781493413,
+ "min_feedback_forward_cosine": 0.9997234344482422
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 53958,
+ "train_examples": 50000,
+ "train_loss": 0.03052342778682709
+ },
+ {
+ "epoch": 139,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023538513109087944,
+ "mean_feedback_forward_cosine": 0.9998464898059243,
+ "mean_feedback_forward_relative_error": 0.017209697098128106,
+ "min_feedback_forward_cosine": 0.9997235536575317
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 54349,
+ "train_examples": 50000,
+ "train_loss": 0.02979899444937706
+ },
+ {
+ "epoch": 140,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023532183840870857,
+ "mean_feedback_forward_cosine": 0.9998465745072616,
+ "mean_feedback_forward_relative_error": 0.017202494911065225,
+ "min_feedback_forward_cosine": 0.9997236728668213
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 54740,
+ "train_examples": 50000,
+ "train_loss": 0.030423043351769447
+ },
+ {
+ "epoch": 141,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023526586592197418,
+ "mean_feedback_forward_cosine": 0.9998467282245034,
+ "mean_feedback_forward_relative_error": 0.017195927150743574,
+ "min_feedback_forward_cosine": 0.9997238516807556
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 55131,
+ "train_examples": 50000,
+ "train_loss": 0.02794100591301918
+ },
+ {
+ "epoch": 142,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023520344868302345,
+ "mean_feedback_forward_cosine": 0.9998468192000138,
+ "mean_feedback_forward_relative_error": 0.01718891215951819,
+ "min_feedback_forward_cosine": 0.9997240304946899
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 55522,
+ "train_examples": 50000,
+ "train_loss": 0.028760734344720842
+ },
+ {
+ "epoch": 143,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02351364493370056,
+ "mean_feedback_forward_cosine": 0.9998469164496974,
+ "mean_feedback_forward_relative_error": 0.017181518683700187,
+ "min_feedback_forward_cosine": 0.9997240900993347
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 55913,
+ "train_examples": 50000,
+ "train_loss": 0.029662667474746704
+ },
+ {
+ "epoch": 144,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02350768819451332,
+ "mean_feedback_forward_cosine": 0.9998470325219003,
+ "mean_feedback_forward_relative_error": 0.017174601922497937,
+ "min_feedback_forward_cosine": 0.9997241497039795
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 56304,
+ "train_examples": 50000,
+ "train_loss": 0.027351453642845153
+ },
+ {
+ "epoch": 145,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023501817137002945,
+ "mean_feedback_forward_cosine": 0.9998471360457571,
+ "mean_feedback_forward_relative_error": 0.01716774528061873,
+ "min_feedback_forward_cosine": 0.9997243881225586
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 56695,
+ "train_examples": 50000,
+ "train_loss": 0.02682767370223999
+ },
+ {
+ "epoch": 146,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023496165871620178,
+ "mean_feedback_forward_cosine": 0.9998472364325273,
+ "mean_feedback_forward_relative_error": 0.017161092598383363,
+ "min_feedback_forward_cosine": 0.9997245073318481
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 57086,
+ "train_examples": 50000,
+ "train_loss": 0.025234533749222757
+ },
+ {
+ "epoch": 147,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023490233346819878,
+ "mean_feedback_forward_cosine": 0.9998473430934706,
+ "mean_feedback_forward_relative_error": 0.01715407116142543,
+ "min_feedback_forward_cosine": 0.9997247457504272
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 57477,
+ "train_examples": 50000,
+ "train_loss": 0.025619760848283768
+ },
+ {
+ "epoch": 148,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02348383143544197,
+ "mean_feedback_forward_cosine": 0.9998474591656735,
+ "mean_feedback_forward_relative_error": 0.017146862062968705,
+ "min_feedback_forward_cosine": 0.999724805355072
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 57868,
+ "train_examples": 50000,
+ "train_loss": 0.027091085720062257
+ },
+ {
+ "epoch": 149,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02347722090780735,
+ "mean_feedback_forward_cosine": 0.9998475877862227,
+ "mean_feedback_forward_relative_error": 0.017139346044706672,
+ "min_feedback_forward_cosine": 0.9997249841690063
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 58259,
+ "train_examples": 50000,
+ "train_loss": 0.02594793123960495
+ },
+ {
+ "epoch": 150,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023470649495720863,
+ "mean_feedback_forward_cosine": 0.9998477038584257,
+ "mean_feedback_forward_relative_error": 0.01713178399950266,
+ "min_feedback_forward_cosine": 0.9997252225875854
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 58650,
+ "train_examples": 50000,
+ "train_loss": 0.025382003281712532
+ },
+ {
+ "epoch": 151,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023470569401979446,
+ "mean_feedback_forward_cosine": 0.9998476818988198,
+ "mean_feedback_forward_relative_error": 0.017131598127123556,
+ "min_feedback_forward_cosine": 0.9997251629829407
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 59041,
+ "train_examples": 50000,
+ "train_loss": 0.022738278942108153
+ },
+ {
+ "epoch": 152,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023470519110560417,
+ "mean_feedback_forward_cosine": 0.9998477038584257,
+ "mean_feedback_forward_relative_error": 0.017131424729565258,
+ "min_feedback_forward_cosine": 0.9997252225875854
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 59432,
+ "train_examples": 50000,
+ "train_loss": 0.02032180576354265
+ },
+ {
+ "epoch": 153,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023470468819141388,
+ "mean_feedback_forward_cosine": 0.9998477352292914,
+ "mean_feedback_forward_relative_error": 0.017131258414960222,
+ "min_feedback_forward_cosine": 0.9997251033782959
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 59823,
+ "train_examples": 50000,
+ "train_loss": 0.02087564205497503
+ },
+ {
+ "epoch": 154,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023470422253012657,
+ "mean_feedback_forward_cosine": 0.9998477132696855,
+ "mean_feedback_forward_relative_error": 0.017131087737844178,
+ "min_feedback_forward_cosine": 0.9997251033782959
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 60214,
+ "train_examples": 50000,
+ "train_loss": 0.020675865701436997
+ },
+ {
+ "epoch": 155,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023470373824238777,
+ "mean_feedback_forward_cosine": 0.999847694447166,
+ "mean_feedback_forward_relative_error": 0.017130916791134758,
+ "min_feedback_forward_cosine": 0.9997250437736511
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 60605,
+ "train_examples": 50000,
+ "train_loss": 0.019594330956339837
+ },
+ {
+ "epoch": 156,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023470336571335793,
+ "mean_feedback_forward_cosine": 0.9998477352292914,
+ "mean_feedback_forward_relative_error": 0.01713075032947879,
+ "min_feedback_forward_cosine": 0.9997252225875854
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 60996,
+ "train_examples": 50000,
+ "train_loss": 0.01812518547475338
+ },
+ {
+ "epoch": 157,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02347029559314251,
+ "mean_feedback_forward_cosine": 0.999847694447166,
+ "mean_feedback_forward_relative_error": 0.017130583696263402,
+ "min_feedback_forward_cosine": 0.9997251629829407
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 61387,
+ "train_examples": 50000,
+ "train_loss": 0.019705076397061348
+ },
+ {
+ "epoch": 158,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02347024716436863,
+ "mean_feedback_forward_cosine": 0.9998477101325989,
+ "mean_feedback_forward_relative_error": 0.017130416205250902,
+ "min_feedback_forward_cosine": 0.9997251033782959
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 61778,
+ "train_examples": 50000,
+ "train_loss": 0.01849426058202982
+ },
+ {
+ "epoch": 159,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023470209911465645,
+ "mean_feedback_forward_cosine": 0.9998477164067721,
+ "mean_feedback_forward_relative_error": 0.017130251777799504,
+ "min_feedback_forward_cosine": 0.9997252225875854
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 62169,
+ "train_examples": 50000,
+ "train_loss": 0.0178389661103487
+ },
+ {
+ "epoch": 160,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023470165207982063,
+ "mean_feedback_forward_cosine": 0.999847738366378,
+ "mean_feedback_forward_relative_error": 0.017130089433569657,
+ "min_feedback_forward_cosine": 0.9997251033782959
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 62560,
+ "train_examples": 50000,
+ "train_loss": 0.018782516342252493
+ },
+ {
+ "epoch": 161,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023470129817724228,
+ "mean_feedback_forward_cosine": 0.9998477195438585,
+ "mean_feedback_forward_relative_error": 0.01712993240768188,
+ "min_feedback_forward_cosine": 0.9997252225875854
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 62951,
+ "train_examples": 50000,
+ "train_loss": 0.017044084163606168
+ },
+ {
+ "epoch": 162,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023470088839530945,
+ "mean_feedback_forward_cosine": 0.9998477007213392,
+ "mean_feedback_forward_relative_error": 0.017129765897008934,
+ "min_feedback_forward_cosine": 0.9997251033782959
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 63342,
+ "train_examples": 50000,
+ "train_loss": 0.018252893408238886
+ },
+ {
+ "epoch": 163,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023470044136047363,
+ "mean_feedback_forward_cosine": 0.9998477226809451,
+ "mean_feedback_forward_relative_error": 0.017129603969423396,
+ "min_feedback_forward_cosine": 0.9997252225875854
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 63733,
+ "train_examples": 50000,
+ "train_loss": 0.018814660705327987
+ },
+ {
+ "epoch": 164,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023469991981983185,
+ "mean_feedback_forward_cosine": 0.9998477226809451,
+ "mean_feedback_forward_relative_error": 0.017129438316547556,
+ "min_feedback_forward_cosine": 0.9997252225875854
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 64124,
+ "train_examples": 50000,
+ "train_loss": 0.01828828042805195
+ },
+ {
+ "epoch": 165,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023469949141144753,
+ "mean_feedback_forward_cosine": 0.9998477101325989,
+ "mean_feedback_forward_relative_error": 0.01712927371753674,
+ "min_feedback_forward_cosine": 0.9997252225875854
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 64515,
+ "train_examples": 50000,
+ "train_loss": 0.017494479485154153
+ },
+ {
+ "epoch": 166,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02346991002559662,
+ "mean_feedback_forward_cosine": 0.9998477038584257,
+ "mean_feedback_forward_relative_error": 0.017129111446832355,
+ "min_feedback_forward_cosine": 0.9997251033782959
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 64906,
+ "train_examples": 50000,
+ "train_loss": 0.016828056658804418
+ },
+ {
+ "epoch": 167,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023469865322113037,
+ "mean_feedback_forward_cosine": 0.9998477195438585,
+ "mean_feedback_forward_relative_error": 0.017128946921347005,
+ "min_feedback_forward_cosine": 0.9997251629829407
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 65297,
+ "train_examples": 50000,
+ "train_loss": 0.01927753235697746
+ },
+ {
+ "epoch": 168,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023469828069210052,
+ "mean_feedback_forward_cosine": 0.9998477571888974,
+ "mean_feedback_forward_relative_error": 0.017128781072403256,
+ "min_feedback_forward_cosine": 0.9997252225875854
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 65688,
+ "train_examples": 50000,
+ "train_loss": 0.015929944035708903
+ },
+ {
+ "epoch": 169,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02346978150308132,
+ "mean_feedback_forward_cosine": 0.9998477352292914,
+ "mean_feedback_forward_relative_error": 0.017128623678888146,
+ "min_feedback_forward_cosine": 0.9997251629829407
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 66079,
+ "train_examples": 50000,
+ "train_loss": 0.017925133748054503
+ },
+ {
+ "epoch": 170,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023469744250178337,
+ "mean_feedback_forward_cosine": 0.9998477320922049,
+ "mean_feedback_forward_relative_error": 0.017128463810015666,
+ "min_feedback_forward_cosine": 0.9997251629829407
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 66470,
+ "train_examples": 50000,
+ "train_loss": 0.016547368803024293
+ },
+ {
+ "epoch": 171,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023469693958759308,
+ "mean_feedback_forward_cosine": 0.9998477477776376,
+ "mean_feedback_forward_relative_error": 0.017128305215584606,
+ "min_feedback_forward_cosine": 0.9997251033782959
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 66861,
+ "train_examples": 50000,
+ "train_loss": 0.018028961103260518
+ },
+ {
+ "epoch": 172,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023469647392630577,
+ "mean_feedback_forward_cosine": 0.9998477446405512,
+ "mean_feedback_forward_relative_error": 0.017128151130715485,
+ "min_feedback_forward_cosine": 0.9997251033782959
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 67252,
+ "train_examples": 50000,
+ "train_loss": 0.016297293134331703
+ },
+ {
+ "epoch": 173,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023469604551792145,
+ "mean_feedback_forward_cosine": 0.9998477164067721,
+ "mean_feedback_forward_relative_error": 0.017127990845198695,
+ "min_feedback_forward_cosine": 0.9997252225875854
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 67643,
+ "train_examples": 50000,
+ "train_loss": 0.016068762745261193
+ },
+ {
+ "epoch": 174,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02346956357359886,
+ "mean_feedback_forward_cosine": 0.9998477320922049,
+ "mean_feedback_forward_relative_error": 0.01712783293700532,
+ "min_feedback_forward_cosine": 0.9997251629829407
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 68034,
+ "train_examples": 50000,
+ "train_loss": 0.017072581720650196
+ },
+ {
+ "epoch": 175,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02346952259540558,
+ "mean_feedback_forward_cosine": 0.999847738366378,
+ "mean_feedback_forward_relative_error": 0.0171276840969528,
+ "min_feedback_forward_cosine": 0.9997251033782959
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 68425,
+ "train_examples": 50000,
+ "train_loss": 0.016299841020703316
+ },
+ {
+ "epoch": 176,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023469489067792892,
+ "mean_feedback_forward_cosine": 0.9998477571888974,
+ "mean_feedback_forward_relative_error": 0.017127525625064186,
+ "min_feedback_forward_cosine": 0.9997251629829407
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 68816,
+ "train_examples": 50000,
+ "train_loss": 0.016555723536014557
+ },
+ {
+ "epoch": 177,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023469455540180206,
+ "mean_feedback_forward_cosine": 0.9998477571888974,
+ "mean_feedback_forward_relative_error": 0.01712736747178592,
+ "min_feedback_forward_cosine": 0.9997252225875854
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 69207,
+ "train_examples": 50000,
+ "train_loss": 0.01746484018266201
+ },
+ {
+ "epoch": 178,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023469412699341774,
+ "mean_feedback_forward_cosine": 0.9998477226809451,
+ "mean_feedback_forward_relative_error": 0.01712720377958919,
+ "min_feedback_forward_cosine": 0.9997251629829407
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 69598,
+ "train_examples": 50000,
+ "train_loss": 0.01591426109135151
+ },
+ {
+ "epoch": 179,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023469367995858192,
+ "mean_feedback_forward_cosine": 0.9998477164067721,
+ "mean_feedback_forward_relative_error": 0.017127045871395814,
+ "min_feedback_forward_cosine": 0.9997251629829407
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 69989,
+ "train_examples": 50000,
+ "train_loss": 0.016877433751672506
+ },
+ {
+ "epoch": 180,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023469330742955208,
+ "mean_feedback_forward_cosine": 0.999847738366378,
+ "mean_feedback_forward_relative_error": 0.017126893502120908,
+ "min_feedback_forward_cosine": 0.9997251033782959
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 70380,
+ "train_examples": 50000,
+ "train_loss": 0.015805002427697182
+ },
+ {
+ "epoch": 181,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023469295352697372,
+ "mean_feedback_forward_cosine": 0.9998477634630705,
+ "mean_feedback_forward_relative_error": 0.017126741941626136,
+ "min_feedback_forward_cosine": 0.9997252225875854
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 70771,
+ "train_examples": 50000,
+ "train_loss": 0.015956093671619893
+ },
+ {
+ "epoch": 182,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02346925064921379,
+ "mean_feedback_forward_cosine": 0.9998477320922049,
+ "mean_feedback_forward_relative_error": 0.01712658337170356,
+ "min_feedback_forward_cosine": 0.9997250437736511
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 71162,
+ "train_examples": 50000,
+ "train_loss": 0.016783491440415382
+ },
+ {
+ "epoch": 183,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023469211533665657,
+ "mean_feedback_forward_cosine": 0.9998477320922049,
+ "mean_feedback_forward_relative_error": 0.01712643097792017,
+ "min_feedback_forward_cosine": 0.9997252225875854
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 71553,
+ "train_examples": 50000,
+ "train_loss": 0.016568301746845246
+ },
+ {
+ "epoch": 184,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023469166830182076,
+ "mean_feedback_forward_cosine": 0.9998477571888974,
+ "mean_feedback_forward_relative_error": 0.017126277089118958,
+ "min_feedback_forward_cosine": 0.9997251629829407
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 71944,
+ "train_examples": 50000,
+ "train_loss": 0.016961991252303122
+ },
+ {
+ "epoch": 185,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02346912771463394,
+ "mean_feedback_forward_cosine": 0.9998477446405512,
+ "mean_feedback_forward_relative_error": 0.017126121239638643,
+ "min_feedback_forward_cosine": 0.9997252225875854
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 72335,
+ "train_examples": 50000,
+ "train_loss": 0.016554582388401032
+ },
+ {
+ "epoch": 186,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02346908301115036,
+ "mean_feedback_forward_cosine": 0.9998477477776376,
+ "mean_feedback_forward_relative_error": 0.017125967963549652,
+ "min_feedback_forward_cosine": 0.9997251629829407
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 72726,
+ "train_examples": 50000,
+ "train_loss": 0.017224053777456283
+ },
+ {
+ "epoch": 187,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02346903830766678,
+ "mean_feedback_forward_cosine": 0.9998477791485033,
+ "mean_feedback_forward_relative_error": 0.017125815030579503,
+ "min_feedback_forward_cosine": 0.9997252225875854
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 73117,
+ "train_examples": 50000,
+ "train_loss": 0.016826375187635422
+ },
+ {
+ "epoch": 188,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023468997329473495,
+ "mean_feedback_forward_cosine": 0.9998477477776376,
+ "mean_feedback_forward_relative_error": 0.017125658592895457,
+ "min_feedback_forward_cosine": 0.9997251629829407
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 73508,
+ "train_examples": 50000,
+ "train_loss": 0.016274484421610832
+ },
+ {
+ "epoch": 189,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023468952625989914,
+ "mean_feedback_forward_cosine": 0.9998477477776376,
+ "mean_feedback_forward_relative_error": 0.017125497155479695,
+ "min_feedback_forward_cosine": 0.9997251629829407
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 73899,
+ "train_examples": 50000,
+ "train_loss": 0.016347632362246515
+ },
+ {
+ "epoch": 190,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023468919098377228,
+ "mean_feedback_forward_cosine": 0.9998477697372437,
+ "mean_feedback_forward_relative_error": 0.017125348119359268,
+ "min_feedback_forward_cosine": 0.9997252821922302
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 74290,
+ "train_examples": 50000,
+ "train_loss": 0.015300385817289352
+ },
+ {
+ "epoch": 191,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023468874394893646,
+ "mean_feedback_forward_cosine": 0.9998477760114168,
+ "mean_feedback_forward_relative_error": 0.017125192171844997,
+ "min_feedback_forward_cosine": 0.9997252821922302
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 74681,
+ "train_examples": 50000,
+ "train_loss": 0.015139786813855171
+ },
+ {
+ "epoch": 192,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02346884272992611,
+ "mean_feedback_forward_cosine": 0.9998477728743302,
+ "mean_feedback_forward_relative_error": 0.017125040096671956,
+ "min_feedback_forward_cosine": 0.9997251629829407
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 75072,
+ "train_examples": 50000,
+ "train_loss": 0.01534516973376274
+ },
+ {
+ "epoch": 193,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023468801751732826,
+ "mean_feedback_forward_cosine": 0.9998477415034646,
+ "mean_feedback_forward_relative_error": 0.01712488878126207,
+ "min_feedback_forward_cosine": 0.9997252225875854
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 75463,
+ "train_examples": 50000,
+ "train_loss": 0.015045764645934104
+ },
+ {
+ "epoch": 194,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02346876822412014,
+ "mean_feedback_forward_cosine": 0.9998477446405512,
+ "mean_feedback_forward_relative_error": 0.01712474187738017,
+ "min_feedback_forward_cosine": 0.9997252225875854
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 75854,
+ "train_examples": 50000,
+ "train_loss": 0.014863526417016984
+ },
+ {
+ "epoch": 195,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023468727245926857,
+ "mean_feedback_forward_cosine": 0.9998477634630705,
+ "mean_feedback_forward_relative_error": 0.017124591591326815,
+ "min_feedback_forward_cosine": 0.9997252225875854
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 76245,
+ "train_examples": 50000,
+ "train_loss": 0.015969047774076463
+ },
+ {
+ "epoch": 196,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023468689993023872,
+ "mean_feedback_forward_cosine": 0.9998477571888974,
+ "mean_feedback_forward_relative_error": 0.017124441623883813,
+ "min_feedback_forward_cosine": 0.9997252225875854
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 76636,
+ "train_examples": 50000,
+ "train_loss": 0.015492048771381378
+ },
+ {
+ "epoch": 197,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023468639701604843,
+ "mean_feedback_forward_cosine": 0.9998477509147242,
+ "mean_feedback_forward_relative_error": 0.017124289034032507,
+ "min_feedback_forward_cosine": 0.9997252225875854
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 77027,
+ "train_examples": 50000,
+ "train_loss": 0.01570508582800627
+ },
+ {
+ "epoch": 198,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02346859872341156,
+ "mean_feedback_forward_cosine": 0.9998477320922049,
+ "mean_feedback_forward_relative_error": 0.01712413610106236,
+ "min_feedback_forward_cosine": 0.9997251033782959
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 77418,
+ "train_examples": 50000,
+ "train_loss": 0.015932950350940226
+ },
+ {
+ "epoch": 199,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023468557745218277,
+ "mean_feedback_forward_cosine": 0.9998477571888974,
+ "mean_feedback_forward_relative_error": 0.017123981133887644,
+ "min_feedback_forward_cosine": 0.9997252225875854
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 77809,
+ "train_examples": 50000,
+ "train_loss": 0.015606247376203537
+ },
+ {
+ "epoch": 200,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023468511179089546,
+ "mean_feedback_forward_cosine": 0.999847760325984,
+ "mean_feedback_forward_relative_error": 0.017123831485054995,
+ "min_feedback_forward_cosine": 0.9997251033782959
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 78200,
+ "train_examples": 50000,
+ "train_loss": 0.015239414473474025
+ }
+ ],
+ "evaluation_protocol": {
+ "test_evaluations": 1,
+ "test_used_for_selection": false,
+ "validation_evaluations": 0
+ },
+ "final": {
+ "accuracy": 0.9155,
+ "epoch": 200,
+ "evaluation_split": "test",
+ "finite": true,
+ "loss": 0.41212791137695315,
+ "step": 78200
+ },
+ "hardware": {
+ "cuda_device_name": "NVIDIA GeForce GTX 1080",
+ "cuda_visible_devices": "7",
+ "device": "cuda",
+ "device_total_memory_bytes": 8507949056,
+ "peak_memory_allocated_bytes": 2162265600,
+ "peak_memory_reserved_bytes": 2864709632,
+ "torch_version": "2.3.1+cu118"
+ },
+ "protocol_family": "oral_a_cifar_local_resnet_development",
+ "provenance": {
+ "git_commit": "0008f2cd96c06bea98f43560867d93967c597711",
+ "git_tracked_dirty": false
+ },
+ "schema_version": 1,
+ "split": {
+ "cifar_source_files": [
+ {
+ "bytes": 31035704,
+ "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_1",
+ "sha256": "54636561a3ce25bd3e19253c6b0d8538147b0ae398331ac4a2d86c6d987368cd"
+ },
+ {
+ "bytes": 31035320,
+ "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_2",
+ "sha256": "766b2cef9fbc745cf056b3152224f7cf77163b330ea9a15f9392beb8b89bc5a8"
+ },
+ {
+ "bytes": 31035999,
+ "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_3",
+ "sha256": "0f00d98ebfb30b3ec0ad19f9756dc2630b89003e10525f5e148445e82aa6a1f9"
+ },
+ {
+ "bytes": 31035696,
+ "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_4",
+ "sha256": "3f7bb240661948b8f4d53e36ec720d8306f5668bd0071dcb4e6c947f78e9682b"
+ },
+ {
+ "bytes": 31035623,
+ "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_5",
+ "sha256": "d91802434d8376bbaeeadf58a737e3a1b12ac839077e931237e0dcd43adcb154"
+ },
+ {
+ "bytes": 31035526,
+ "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/test_batch",
+ "sha256": "f53d8d457504f7cff4ea9e021afcf0e0ad8e24a91f3fc42091b8adef61157831"
+ }
+ ],
+ "dataset": "cifar10",
+ "evaluation_split": "test",
+ "input_layout": "NCHW",
+ "input_shape": [
+ 3,
+ 32,
+ 32
+ ],
+ "loader_seed": 11,
+ "normalization_mean": [
+ 0.49140000343322754,
+ 0.4821999967098236,
+ 0.4465000033378601
+ ],
+ "normalization_std": [
+ 0.24699999392032623,
+ 0.2434999942779541,
+ 0.26159998774528503
+ ],
+ "split_from_training_only": true,
+ "split_seed": null,
+ "test_examples": 10000,
+ "train_examples": 50000,
+ "training_augmentation": "random_crop_32_padding_4_zero_then_horizontal_flip_p0.5",
+ "validation_class_counts": {},
+ "validation_examples": 0,
+ "validation_index_sha256": null
+ },
+ "timing": {
+ "apical_warmup_wall_s": 0.0,
+ "evaluation_wall_s": 0.7048258781433105,
+ "mirror_warmup_wall_s": 0.0,
+ "predictor_warmup_wall_s": 0.0,
+ "timing_excludes_data_loading_hashing_and_model_construction": true,
+ "total_timed_wall_s": 6205.603518009186,
+ "train_wall_s": 6203.436691999435
+ },
+ "work": {
+ "apical_macs_per_example": 40108672,
+ "causal_scalar_observations": 0,
+ "components": {
+ "apical_projection_macs": 401086720000000,
+ "apical_regression_macs": 0,
+ "bp_reverse_macs_estimate": 0,
+ "kp_reciprocal_correlation_macs": 401086720000000,
+ "local_weight_correlation_macs": 405510400000000,
+ "mirror_feedback_prediction_macs": 0,
+ "mirror_local_correlation_macs": 0,
+ "mirror_response_macs": 0,
+ "ordinary_forward_macs": 405510400000000,
+ "perturbation_forward_macs": 0,
+ "warmup_clean_forward_macs": 0
+ },
+ "definition": "multiply-accumulates in conv/linear maps; one local weight correlation equals one forward-weight MAC count; BP reverse is estimated as one weight-gradient plus one activation-gradient convolution per forward convolution; mixed-traffic/predictor elementwise arithmetic is reported as a conservative operation estimate separately and is not folded into MACs",
+ "elementwise_operations_estimate": 0,
+ "forward_macs_per_example": 40551040,
+ "logical_batch_loss_queries": 0,
+ "mirror_probe_examples": 0,
+ "mirror_readout_probe_examples": 0,
+ "neutral_projection_observations": 0,
+ "per_example_cross_entropy_terms": 0,
+ "total_clean_forward_examples": 10000000,
+ "total_forward_equivalent_examples": 10000000,
+ "total_macs_estimate": 1613194240000000
+ }
+}