summaryrefslogtreecommitdiff
path: root/results/kp_dynamic_projection_confirmation/seed12_clean_kp.json
diff options
context:
space:
mode:
Diffstat (limited to 'results/kp_dynamic_projection_confirmation/seed12_clean_kp.json')
-rw-r--r--results/kp_dynamic_projection_confirmation/seed12_clean_kp.json3258
1 files changed, 3258 insertions, 0 deletions
diff --git a/results/kp_dynamic_projection_confirmation/seed12_clean_kp.json b/results/kp_dynamic_projection_confirmation/seed12_clean_kp.json
new file mode 100644
index 0000000..80bd05a
--- /dev/null
+++ b/results/kp_dynamic_projection_confirmation/seed12_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": 12,
+ "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/seed12_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": 12,
+ "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": 0.9996837178866068,
+ "feedback_forward_cosine": [
+ 0.9998112320899963,
+ 0.9998036026954651,
+ 0.9998302459716797,
+ 0.9998162984848022,
+ 0.9998250007629395,
+ 0.9997634887695312,
+ 0.9998230934143066,
+ 0.9998724460601807,
+ 0.9998655915260315,
+ 0.9998391270637512,
+ 0.9998552799224854,
+ 0.9998058080673218,
+ 0.9998328685760498,
+ 0.9999038577079773,
+ 0.9998883008956909,
+ 0.9998690485954285,
+ 0.999883234500885,
+ 0.9997214078903198,
+ 0.9999873638153076
+ ],
+ "feedback_forward_norm_ratio": [
+ 0.9997771978378296,
+ 1.0001469850540161,
+ 0.9989446401596069,
+ 1.000712275505066,
+ 0.9991862177848816,
+ 1.0010864734649658,
+ 0.9990431666374207,
+ 1.0001176595687866,
+ 0.9995997548103333,
+ 1.00063157081604,
+ 0.9998372793197632,
+ 1.0005996227264404,
+ 0.9999316334724426,
+ 1.0002691745758057,
+ 0.9999130964279175,
+ 1.0005886554718018,
+ 0.9999468326568604,
+ 1.0006757974624634,
+ 1.0023469924926758
+ ],
+ "feedback_forward_relative_error": [
+ 0.019424350932240486,
+ 0.019819237291812897,
+ 0.018446823582053185,
+ 0.019186608493328094,
+ 0.018720431253314018,
+ 0.021783743053674698,
+ 0.018828051164746284,
+ 0.015970267355442047,
+ 0.01640067622065544,
+ 0.017956888303160667,
+ 0.01700727455317974,
+ 0.01971874199807644,
+ 0.018282445147633553,
+ 0.013876795768737793,
+ 0.01494023110717535,
+ 0.016201160848140717,
+ 0.015274452976882458,
+ 0.02362348884344101,
+ 0.005561360623687506
+ ],
+ "innovation_negative_gradient_cosine": [
+ 0.9997078776359558,
+ 0.9996459484100342,
+ 0.9997032880783081,
+ 0.9996677041053772,
+ 0.9997084140777588,
+ 0.999669075012207,
+ 0.9997341632843018,
+ 0.9997588992118835,
+ 0.9997591972351074,
+ 0.9997448921203613,
+ 0.9997714161872864,
+ 0.999757707118988,
+ 0.9997740387916565,
+ 0.999829888343811,
+ 0.9998507499694824,
+ 0.9998759627342224,
+ 0.9999287128448486,
+ 0.999973475933075,
+ 0.7562967538833618
+ ],
+ "mean_feedback_forward_cosine": 0.9998419629900079,
+ "mean_feedback_forward_relative_error": 0.017422264711441177,
+ "normalization_state": "training_batch_stats_without_running_update",
+ "raw_negative_gradient_cosine": [
+ 0.9997078776359558,
+ 0.9996459484100342,
+ 0.9997032880783081,
+ 0.9996677041053772,
+ 0.9997084140777588,
+ 0.999669075012207,
+ 0.9997341632843018,
+ 0.9997588992118835,
+ 0.9997591972351074,
+ 0.9997448921203613,
+ 0.9997714161872864,
+ 0.999757707118988,
+ 0.9997740387916565,
+ 0.999829888343811,
+ 0.9998507499694824,
+ 0.9998759627342224,
+ 0.9999287128448486,
+ 0.999973475933075,
+ 0.7562967538833618
+ ],
+ "teaching_negative_gradient_cosine": [
+ 0.9997078776359558,
+ 0.9996459484100342,
+ 0.9997032880783081,
+ 0.9996677041053772,
+ 0.9997084140777588,
+ 0.999669075012207,
+ 0.9997341632843018,
+ 0.9997588992118835,
+ 0.9997591972351074,
+ 0.9997448921203613,
+ 0.9997714161872864,
+ 0.999757707118988,
+ 0.9997740387916565,
+ 0.999829888343811,
+ 0.9998507499694824,
+ 0.9998759627342224,
+ 0.9999287128448486,
+ 0.999973475933075,
+ 0.7562967538833618
+ ],
+ "wall_s": 0.10664820671081543
+ },
+ "epochs": [
+ {
+ "epoch": 1,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.3357337713241577,
+ "mean_feedback_forward_cosine": 0.2956587334996776,
+ "mean_feedback_forward_relative_error": 1.1827429689859088,
+ "min_feedback_forward_cosine": 0.11035917699337006
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 391,
+ "train_examples": 50000,
+ "train_loss": 1.9535887855911256
+ },
+ {
+ "epoch": 2,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.2760885953903198,
+ "mean_feedback_forward_cosine": 0.4008528252965526,
+ "mean_feedback_forward_relative_error": 1.0911526554509212,
+ "min_feedback_forward_cosine": 0.19288809597492218
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 782,
+ "train_examples": 50000,
+ "train_loss": 1.425792371482849
+ },
+ {
+ "epoch": 3,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.2156856060028076,
+ "mean_feedback_forward_cosine": 0.4969612971732491,
+ "mean_feedback_forward_relative_error": 1.0003732176203477,
+ "min_feedback_forward_cosine": 0.27784961462020874
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 1173,
+ "train_examples": 50000,
+ "train_loss": 1.1101230571365357
+ },
+ {
+ "epoch": 4,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.1704801321029663,
+ "mean_feedback_forward_cosine": 0.5665066602982973,
+ "mean_feedback_forward_relative_error": 0.9287926507623572,
+ "min_feedback_forward_cosine": 0.33664584159851074
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 1564,
+ "train_examples": 50000,
+ "train_loss": 0.9170437034416199
+ },
+ {
+ "epoch": 5,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.127273678779602,
+ "mean_feedback_forward_cosine": 0.6199988314979955,
+ "mean_feedback_forward_relative_error": 0.8690415071813684,
+ "min_feedback_forward_cosine": 0.38803422451019287
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 1955,
+ "train_examples": 50000,
+ "train_loss": 0.8006134103012085
+ },
+ {
+ "epoch": 6,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.0858829021453857,
+ "mean_feedback_forward_cosine": 0.6622635737845772,
+ "mean_feedback_forward_relative_error": 0.8191626479751185,
+ "min_feedback_forward_cosine": 0.43329018354415894
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 2346,
+ "train_examples": 50000,
+ "train_loss": 0.7192629204177856
+ },
+ {
+ "epoch": 7,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.0461058616638184,
+ "mean_feedback_forward_cosine": 0.6982236128104361,
+ "mean_feedback_forward_relative_error": 0.7741572872588509,
+ "min_feedback_forward_cosine": 0.47564470767974854
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 2737,
+ "train_examples": 50000,
+ "train_loss": 0.6592539573287964
+ },
+ {
+ "epoch": 8,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.0087120532989502,
+ "mean_feedback_forward_cosine": 0.7292329041581405,
+ "mean_feedback_forward_relative_error": 0.7327324183363664,
+ "min_feedback_forward_cosine": 0.5136085748672485
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 3128,
+ "train_examples": 50000,
+ "train_loss": 0.620042862815857
+ },
+ {
+ "epoch": 9,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.9720722436904907,
+ "mean_feedback_forward_cosine": 0.755608175930224,
+ "mean_feedback_forward_relative_error": 0.6957467841474634,
+ "min_feedback_forward_cosine": 0.5488060116767883
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 3519,
+ "train_examples": 50000,
+ "train_loss": 0.5835777080154418
+ },
+ {
+ "epoch": 10,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.9355390667915344,
+ "mean_feedback_forward_cosine": 0.778680471997512,
+ "mean_feedback_forward_relative_error": 0.6615886656861556,
+ "min_feedback_forward_cosine": 0.582585334777832
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 3910,
+ "train_examples": 50000,
+ "train_loss": 0.5475232218837738
+ },
+ {
+ "epoch": 11,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.9007604122161865,
+ "mean_feedback_forward_cosine": 0.7992428007878756,
+ "mean_feedback_forward_relative_error": 0.6297440058306644,
+ "min_feedback_forward_cosine": 0.6133478879928589
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 4301,
+ "train_examples": 50000,
+ "train_loss": 0.5219385059165954
+ },
+ {
+ "epoch": 12,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.8674523830413818,
+ "mean_feedback_forward_cosine": 0.8169901778823451,
+ "mean_feedback_forward_relative_error": 0.6008071617076272,
+ "min_feedback_forward_cosine": 0.6414960026741028
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 4692,
+ "train_examples": 50000,
+ "train_loss": 0.5009035326194763
+ },
+ {
+ "epoch": 13,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.8340794444084167,
+ "mean_feedback_forward_cosine": 0.8333831617706701,
+ "mean_feedback_forward_relative_error": 0.5730628041844619,
+ "min_feedback_forward_cosine": 0.6686401963233948
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 5083,
+ "train_examples": 50000,
+ "train_loss": 0.4774324990940094
+ },
+ {
+ "epoch": 14,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.8040633797645569,
+ "mean_feedback_forward_cosine": 0.8475955881570515,
+ "mean_feedback_forward_relative_error": 0.5479318938757244,
+ "min_feedback_forward_cosine": 0.6920316219329834
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 5474,
+ "train_examples": 50000,
+ "train_loss": 0.4525343029975891
+ },
+ {
+ "epoch": 15,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.7733475565910339,
+ "mean_feedback_forward_cosine": 0.8608487468016776,
+ "mean_feedback_forward_relative_error": 0.5234691524191907,
+ "min_feedback_forward_cosine": 0.7149572372436523
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 5865,
+ "train_examples": 50000,
+ "train_loss": 0.44091602117538453
+ },
+ {
+ "epoch": 16,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.7440672516822815,
+ "mean_feedback_forward_cosine": 0.8724623197003415,
+ "mean_feedback_forward_relative_error": 0.5008108506077215,
+ "min_feedback_forward_cosine": 0.7361483573913574
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 6256,
+ "train_examples": 50000,
+ "train_loss": 0.427868346118927
+ },
+ {
+ "epoch": 17,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.7161498069763184,
+ "mean_feedback_forward_cosine": 0.8828342086390445,
+ "mean_feedback_forward_relative_error": 0.4797582257735102,
+ "min_feedback_forward_cosine": 0.7555732131004333
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 6647,
+ "train_examples": 50000,
+ "train_loss": 0.4159619722270966
+ },
+ {
+ "epoch": 18,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.6886909008026123,
+ "mean_feedback_forward_cosine": 0.8923413973105582,
+ "mean_feedback_forward_relative_error": 0.4595971374135268,
+ "min_feedback_forward_cosine": 0.7737658023834229
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 7038,
+ "train_examples": 50000,
+ "train_loss": 0.39836229312896726
+ },
+ {
+ "epoch": 19,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.6615512371063232,
+ "mean_feedback_forward_cosine": 0.9010440017047682,
+ "mean_feedback_forward_relative_error": 0.4405165267618079,
+ "min_feedback_forward_cosine": 0.7909722328186035
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 7429,
+ "train_examples": 50000,
+ "train_loss": 0.38774474955558774
+ },
+ {
+ "epoch": 20,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.6349936127662659,
+ "mean_feedback_forward_cosine": 0.9091169049865321,
+ "mean_feedback_forward_relative_error": 0.42201937656653554,
+ "min_feedback_forward_cosine": 0.8073540925979614
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 7820,
+ "train_examples": 50000,
+ "train_loss": 0.3837249806308746
+ },
+ {
+ "epoch": 21,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.6095899343490601,
+ "mean_feedback_forward_cosine": 0.916544280554119,
+ "mean_feedback_forward_relative_error": 0.4042697760619615,
+ "min_feedback_forward_cosine": 0.8222854137420654
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 8211,
+ "train_examples": 50000,
+ "train_loss": 0.3678284918117523
+ },
+ {
+ "epoch": 22,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.5848206877708435,
+ "mean_feedback_forward_cosine": 0.9232736637717799,
+ "mean_feedback_forward_relative_error": 0.38750572424185903,
+ "min_feedback_forward_cosine": 0.836258590221405
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 8602,
+ "train_examples": 50000,
+ "train_loss": 0.36831582045555117
+ },
+ {
+ "epoch": 23,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.5614774227142334,
+ "mean_feedback_forward_cosine": 0.9293549939205772,
+ "mean_feedback_forward_relative_error": 0.3717757416398902,
+ "min_feedback_forward_cosine": 0.8488701581954956
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 8993,
+ "train_examples": 50000,
+ "train_loss": 0.3573762542152405
+ },
+ {
+ "epoch": 24,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.5387479066848755,
+ "mean_feedback_forward_cosine": 0.9349046443638048,
+ "mean_feedback_forward_relative_error": 0.3567409209514919,
+ "min_feedback_forward_cosine": 0.8606550693511963
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 9384,
+ "train_examples": 50000,
+ "train_loss": 0.3481489736366272
+ },
+ {
+ "epoch": 25,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.5167993307113647,
+ "mean_feedback_forward_cosine": 0.9400349478972586,
+ "mean_feedback_forward_relative_error": 0.34226871556357336,
+ "min_feedback_forward_cosine": 0.8716477155685425
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 9775,
+ "train_examples": 50000,
+ "train_loss": 0.33956556100845336
+ },
+ {
+ "epoch": 26,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.49593883752822876,
+ "mean_feedback_forward_cosine": 0.9447282427235654,
+ "mean_feedback_forward_relative_error": 0.3285521072776694,
+ "min_feedback_forward_cosine": 0.8816639184951782
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 10166,
+ "train_examples": 50000,
+ "train_loss": 0.33481121618270876
+ },
+ {
+ "epoch": 27,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.4741222560405731,
+ "mean_feedback_forward_cosine": 0.9490944805898165,
+ "mean_feedback_forward_relative_error": 0.315203668255555,
+ "min_feedback_forward_cosine": 0.8916881680488586
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 10557,
+ "train_examples": 50000,
+ "train_loss": 0.3340137974381447
+ },
+ {
+ "epoch": 28,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.4551987946033478,
+ "mean_feedback_forward_cosine": 0.9530173602857088,
+ "mean_feedback_forward_relative_error": 0.30273031953134033,
+ "min_feedback_forward_cosine": 0.9000020027160645
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 10948,
+ "train_examples": 50000,
+ "train_loss": 0.3191308118057251
+ },
+ {
+ "epoch": 29,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.43651825189590454,
+ "mean_feedback_forward_cosine": 0.9566915129360399,
+ "mean_feedback_forward_relative_error": 0.290577570074483,
+ "min_feedback_forward_cosine": 0.9079872369766235
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 11339,
+ "train_examples": 50000,
+ "train_loss": 0.3136642800617218
+ },
+ {
+ "epoch": 30,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.41827085614204407,
+ "mean_feedback_forward_cosine": 0.960089601968464,
+ "mean_feedback_forward_relative_error": 0.27891130666983754,
+ "min_feedback_forward_cosine": 0.915353000164032
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 11730,
+ "train_examples": 50000,
+ "train_loss": 0.3164820732307434
+ },
+ {
+ "epoch": 31,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.40173670649528503,
+ "mean_feedback_forward_cosine": 0.963170842120522,
+ "mean_feedback_forward_relative_error": 0.26793484585849864,
+ "min_feedback_forward_cosine": 0.9218012690544128
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 12121,
+ "train_examples": 50000,
+ "train_loss": 0.30453836922645566
+ },
+ {
+ "epoch": 32,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.38502153754234314,
+ "mean_feedback_forward_cosine": 0.9659909794205114,
+ "mean_feedback_forward_relative_error": 0.2573729816235994,
+ "min_feedback_forward_cosine": 0.9281191229820251
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 12512,
+ "train_examples": 50000,
+ "train_loss": 0.29894664553642275
+ },
+ {
+ "epoch": 33,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.36931300163269043,
+ "mean_feedback_forward_cosine": 0.9685940428784019,
+ "mean_feedback_forward_relative_error": 0.24731895366781637,
+ "min_feedback_forward_cosine": 0.9338076114654541
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 12903,
+ "train_examples": 50000,
+ "train_loss": 0.29957063269615175
+ },
+ {
+ "epoch": 34,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.3545270562171936,
+ "mean_feedback_forward_cosine": 0.9709650968250475,
+ "mean_feedback_forward_relative_error": 0.23778290772124341,
+ "min_feedback_forward_cosine": 0.9388999938964844
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 13294,
+ "train_examples": 50000,
+ "train_loss": 0.28676708374023435
+ },
+ {
+ "epoch": 35,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.3397584557533264,
+ "mean_feedback_forward_cosine": 0.9731669300480893,
+ "mean_feedback_forward_relative_error": 0.22850124616371958,
+ "min_feedback_forward_cosine": 0.9438261985778809
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 13685,
+ "train_examples": 50000,
+ "train_loss": 0.2893641914558411
+ },
+ {
+ "epoch": 36,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.32634857296943665,
+ "mean_feedback_forward_cosine": 0.975176403397008,
+ "mean_feedback_forward_relative_error": 0.21974362393743113,
+ "min_feedback_forward_cosine": 0.9481179118156433
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 14076,
+ "train_examples": 50000,
+ "train_loss": 0.2815560569381714
+ },
+ {
+ "epoch": 37,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.31317803263664246,
+ "mean_feedback_forward_cosine": 0.9770319775531167,
+ "mean_feedback_forward_relative_error": 0.21132852019448028,
+ "min_feedback_forward_cosine": 0.9521644115447998
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 14467,
+ "train_examples": 50000,
+ "train_loss": 0.27776203176498415
+ },
+ {
+ "epoch": 38,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.30006417632102966,
+ "mean_feedback_forward_cosine": 0.9788156936043187,
+ "mean_feedback_forward_relative_error": 0.2029386492151963,
+ "min_feedback_forward_cosine": 0.9560489654541016
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 14858,
+ "train_examples": 50000,
+ "train_loss": 0.2797470482444763
+ },
+ {
+ "epoch": 39,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.2883997857570648,
+ "mean_feedback_forward_cosine": 0.9804035864378277,
+ "mean_feedback_forward_relative_error": 0.1951562425023631,
+ "min_feedback_forward_cosine": 0.959380030632019
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 15249,
+ "train_examples": 50000,
+ "train_loss": 0.270851840467453
+ },
+ {
+ "epoch": 40,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.27658069133758545,
+ "mean_feedback_forward_cosine": 0.9819350179873014,
+ "mean_feedback_forward_relative_error": 0.18736776827197327,
+ "min_feedback_forward_cosine": 0.9625852108001709
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 15640,
+ "train_examples": 50000,
+ "train_loss": 0.2707836451721191
+ },
+ {
+ "epoch": 41,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.265087366104126,
+ "mean_feedback_forward_cosine": 0.9833378980034276,
+ "mean_feedback_forward_relative_error": 0.179935664722794,
+ "min_feedback_forward_cosine": 0.9655967950820923
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 16031,
+ "train_examples": 50000,
+ "train_loss": 0.2751759523200989
+ },
+ {
+ "epoch": 42,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.25412866473197937,
+ "mean_feedback_forward_cosine": 0.9846446043566653,
+ "mean_feedback_forward_relative_error": 0.17271635955885836,
+ "min_feedback_forward_cosine": 0.9683431386947632
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 16422,
+ "train_examples": 50000,
+ "train_loss": 0.2680864411687851
+ },
+ {
+ "epoch": 43,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.24418483674526215,
+ "mean_feedback_forward_cosine": 0.9857830718943947,
+ "mean_feedback_forward_relative_error": 0.16614990053992523,
+ "min_feedback_forward_cosine": 0.9707401394844055
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 16813,
+ "train_examples": 50000,
+ "train_loss": 0.26244630385398865
+ },
+ {
+ "epoch": 44,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.23399662971496582,
+ "mean_feedback_forward_cosine": 0.9868629731630024,
+ "mean_feedback_forward_relative_error": 0.15971490466280988,
+ "min_feedback_forward_cosine": 0.9731128215789795
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 17204,
+ "train_examples": 50000,
+ "train_loss": 0.26145189335107805
+ },
+ {
+ "epoch": 45,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.22435632348060608,
+ "mean_feedback_forward_cosine": 0.9878626246201364,
+ "mean_feedback_forward_relative_error": 0.15350551432684847,
+ "min_feedback_forward_cosine": 0.9752629995346069
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 17595,
+ "train_examples": 50000,
+ "train_loss": 0.25927943958282473
+ },
+ {
+ "epoch": 46,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.2154891937971115,
+ "mean_feedback_forward_cosine": 0.9887609826891046,
+ "mean_feedback_forward_relative_error": 0.1476938493157688,
+ "min_feedback_forward_cosine": 0.9771661162376404
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 17986,
+ "train_examples": 50000,
+ "train_loss": 0.2541186200237274
+ },
+ {
+ "epoch": 47,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.20674291253089905,
+ "mean_feedback_forward_cosine": 0.9896312293253446,
+ "mean_feedback_forward_relative_error": 0.14187976423847048,
+ "min_feedback_forward_cosine": 0.9789839386940002
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 18377,
+ "train_examples": 50000,
+ "train_loss": 0.2508393686246872
+ },
+ {
+ "epoch": 48,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.19846054911613464,
+ "mean_feedback_forward_cosine": 0.990399275955401,
+ "mean_feedback_forward_relative_error": 0.13646776699706128,
+ "min_feedback_forward_cosine": 0.9806078672409058
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 18768,
+ "train_examples": 50000,
+ "train_loss": 0.2510678723812103
+ },
+ {
+ "epoch": 49,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.19029174745082855,
+ "mean_feedback_forward_cosine": 0.9911266910402399,
+ "mean_feedback_forward_relative_error": 0.1311942199735265,
+ "min_feedback_forward_cosine": 0.9821584224700928
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 19159,
+ "train_examples": 50000,
+ "train_loss": 0.24962311874389648
+ },
+ {
+ "epoch": 50,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.18236908316612244,
+ "mean_feedback_forward_cosine": 0.9918002643083271,
+ "mean_feedback_forward_relative_error": 0.12609709309119926,
+ "min_feedback_forward_cosine": 0.9835997819900513
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 19550,
+ "train_examples": 50000,
+ "train_loss": 0.2451121556663513
+ },
+ {
+ "epoch": 51,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.1750110238790512,
+ "mean_feedback_forward_cosine": 0.9924097312124152,
+ "mean_feedback_forward_relative_error": 0.1213228457460278,
+ "min_feedback_forward_cosine": 0.9848860502243042
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 19941,
+ "train_examples": 50000,
+ "train_loss": 0.24124271689414978
+ },
+ {
+ "epoch": 52,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.1679934561252594,
+ "mean_feedback_forward_cosine": 0.992973484491047,
+ "mean_feedback_forward_relative_error": 0.11671906043040126,
+ "min_feedback_forward_cosine": 0.9860671758651733
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 20332,
+ "train_examples": 50000,
+ "train_loss": 0.2436721534538269
+ },
+ {
+ "epoch": 53,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.16107222437858582,
+ "mean_feedback_forward_cosine": 0.9935088973296317,
+ "mean_feedback_forward_relative_error": 0.11218417651559177,
+ "min_feedback_forward_cosine": 0.9871819019317627
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 20723,
+ "train_examples": 50000,
+ "train_loss": 0.24679876781463622
+ },
+ {
+ "epoch": 54,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.1547631323337555,
+ "mean_feedback_forward_cosine": 0.9939876543848138,
+ "mean_feedback_forward_relative_error": 0.10794075088281381,
+ "min_feedback_forward_cosine": 0.9881608486175537
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 21114,
+ "train_examples": 50000,
+ "train_loss": 0.2385181399536133
+ },
+ {
+ "epoch": 55,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.14851979911327362,
+ "mean_feedback_forward_cosine": 0.9944537062393991,
+ "mean_feedback_forward_relative_error": 0.10368357128218601,
+ "min_feedback_forward_cosine": 0.9890934228897095
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 21505,
+ "train_examples": 50000,
+ "train_loss": 0.24382776163101197
+ },
+ {
+ "epoch": 56,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.14255520701408386,
+ "mean_feedback_forward_cosine": 0.994874329943406,
+ "mean_feedback_forward_relative_error": 0.09966899434986867,
+ "min_feedback_forward_cosine": 0.9899448752403259
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 21896,
+ "train_examples": 50000,
+ "train_loss": 0.23560380182266236
+ },
+ {
+ "epoch": 57,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.13663329184055328,
+ "mean_feedback_forward_cosine": 0.9952741547634727,
+ "mean_feedback_forward_relative_error": 0.09569971930039556,
+ "min_feedback_forward_cosine": 0.9907571077346802
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 22287,
+ "train_examples": 50000,
+ "train_loss": 0.23805610733032226
+ },
+ {
+ "epoch": 58,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.13108128309249878,
+ "mean_feedback_forward_cosine": 0.9956361620049727,
+ "mean_feedback_forward_relative_error": 0.09196121990680695,
+ "min_feedback_forward_cosine": 0.9914897680282593
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 22678,
+ "train_examples": 50000,
+ "train_loss": 0.2356652353286743
+ },
+ {
+ "epoch": 59,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.12578646838665009,
+ "mean_feedback_forward_cosine": 0.9959604489175897,
+ "mean_feedback_forward_relative_error": 0.08847149912463992,
+ "min_feedback_forward_cosine": 0.9921623468399048
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 23069,
+ "train_examples": 50000,
+ "train_loss": 0.23241991509914398
+ },
+ {
+ "epoch": 60,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.12085700780153275,
+ "mean_feedback_forward_cosine": 0.9962478217325712,
+ "mean_feedback_forward_relative_error": 0.0852508527275763,
+ "min_feedback_forward_cosine": 0.9927586913108826
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 23460,
+ "train_examples": 50000,
+ "train_loss": 0.23017726940155028
+ },
+ {
+ "epoch": 61,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.11604860424995422,
+ "mean_feedback_forward_cosine": 0.9965282960941917,
+ "mean_feedback_forward_relative_error": 0.08199545074450343,
+ "min_feedback_forward_cosine": 0.9933216571807861
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 23851,
+ "train_examples": 50000,
+ "train_loss": 0.2279142566394806
+ },
+ {
+ "epoch": 62,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.11142109334468842,
+ "mean_feedback_forward_cosine": 0.9967884295865109,
+ "mean_feedback_forward_relative_error": 0.07886457835373126,
+ "min_feedback_forward_cosine": 0.9938427209854126
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 24242,
+ "train_examples": 50000,
+ "train_loss": 0.22619809977531433
+ },
+ {
+ "epoch": 63,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.10686975717544556,
+ "mean_feedback_forward_cosine": 0.9970326831466273,
+ "mean_feedback_forward_relative_error": 0.07581320601074319,
+ "min_feedback_forward_cosine": 0.9943329691886902
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 24633,
+ "train_examples": 50000,
+ "train_loss": 0.22850586309432983
+ },
+ {
+ "epoch": 64,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.10264051705598831,
+ "mean_feedback_forward_cosine": 0.9972573895203439,
+ "mean_feedback_forward_relative_error": 0.07289840908426988,
+ "min_feedback_forward_cosine": 0.9947707653045654
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 25024,
+ "train_examples": 50000,
+ "train_loss": 0.22459484602928162
+ },
+ {
+ "epoch": 65,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.09847520291805267,
+ "mean_feedback_forward_cosine": 0.997464961127231,
+ "mean_feedback_forward_relative_error": 0.07008228382389796,
+ "min_feedback_forward_cosine": 0.995185375213623
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 25415,
+ "train_examples": 50000,
+ "train_loss": 0.2218525476360321
+ },
+ {
+ "epoch": 66,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.09460356831550598,
+ "mean_feedback_forward_cosine": 0.997650212363193,
+ "mean_feedback_forward_relative_error": 0.06746487407700012,
+ "min_feedback_forward_cosine": 0.9955543875694275
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 25806,
+ "train_examples": 50000,
+ "train_loss": 0.221656060962677
+ },
+ {
+ "epoch": 67,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0907011330127716,
+ "mean_feedback_forward_cosine": 0.9978318496754295,
+ "mean_feedback_forward_relative_error": 0.06480307298663415,
+ "min_feedback_forward_cosine": 0.9959129095077515
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 26197,
+ "train_examples": 50000,
+ "train_loss": 0.2231084961462021
+ },
+ {
+ "epoch": 68,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.08713283389806747,
+ "mean_feedback_forward_cosine": 0.997995022096132,
+ "mean_feedback_forward_relative_error": 0.062316504356108214,
+ "min_feedback_forward_cosine": 0.9962259531021118
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 26588,
+ "train_examples": 50000,
+ "train_loss": 0.22055475136756897
+ },
+ {
+ "epoch": 69,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.08364100754261017,
+ "mean_feedback_forward_cosine": 0.9981481432914734,
+ "mean_feedback_forward_relative_error": 0.05989345251337478,
+ "min_feedback_forward_cosine": 0.9965221285820007
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 26979,
+ "train_examples": 50000,
+ "train_loss": 0.21850419645786284
+ },
+ {
+ "epoch": 70,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.080176942050457,
+ "mean_feedback_forward_cosine": 0.9982891521955791,
+ "mean_feedback_forward_relative_error": 0.057557923817320875,
+ "min_feedback_forward_cosine": 0.9968026876449585
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 27370,
+ "train_examples": 50000,
+ "train_loss": 0.22419196217536927
+ },
+ {
+ "epoch": 71,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.07705311477184296,
+ "mean_feedback_forward_cosine": 0.9984161069518641,
+ "mean_feedback_forward_relative_error": 0.055389606540924625,
+ "min_feedback_forward_cosine": 0.9970472455024719
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 27761,
+ "train_examples": 50000,
+ "train_loss": 0.21988929317474365
+ },
+ {
+ "epoch": 72,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.07398562878370285,
+ "mean_feedback_forward_cosine": 0.9985364801005313,
+ "mean_feedback_forward_relative_error": 0.0532407456714856,
+ "min_feedback_forward_cosine": 0.9972772002220154
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 28152,
+ "train_examples": 50000,
+ "train_loss": 0.21685568919181825
+ },
+ {
+ "epoch": 73,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.07109541445970535,
+ "mean_feedback_forward_cosine": 0.9986450138844942,
+ "mean_feedback_forward_relative_error": 0.05122427073748488,
+ "min_feedback_forward_cosine": 0.9974852800369263
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 28543,
+ "train_examples": 50000,
+ "train_loss": 0.21576430081367493
+ },
+ {
+ "epoch": 74,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.06826361268758774,
+ "mean_feedback_forward_cosine": 0.9987454979043258,
+ "mean_feedback_forward_relative_error": 0.04928921859123205,
+ "min_feedback_forward_cosine": 0.9976809024810791
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 28934,
+ "train_examples": 50000,
+ "train_loss": 0.21800933145046233
+ },
+ {
+ "epoch": 75,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.06553558260202408,
+ "mean_feedback_forward_cosine": 0.9988394881549635,
+ "mean_feedback_forward_relative_error": 0.04740385789620249,
+ "min_feedback_forward_cosine": 0.9978623986244202
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 29325,
+ "train_examples": 50000,
+ "train_loss": 0.2126266315174103
+ },
+ {
+ "epoch": 76,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.06292152404785156,
+ "mean_feedback_forward_cosine": 0.9989262568323236,
+ "mean_feedback_forward_relative_error": 0.04560201899393609,
+ "min_feedback_forward_cosine": 0.9980289936065674
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 29716,
+ "train_examples": 50000,
+ "train_loss": 0.21306990696907044
+ },
+ {
+ "epoch": 77,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.060485124588012695,
+ "mean_feedback_forward_cosine": 0.999006167838448,
+ "mean_feedback_forward_relative_error": 0.04386914844967817,
+ "min_feedback_forward_cosine": 0.9981784820556641
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 30107,
+ "train_examples": 50000,
+ "train_loss": 0.2119411366558075
+ },
+ {
+ "epoch": 78,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.05806891247630119,
+ "mean_feedback_forward_cosine": 0.9990817998584948,
+ "mean_feedback_forward_relative_error": 0.042162092010441576,
+ "min_feedback_forward_cosine": 0.9983206391334534
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 30498,
+ "train_examples": 50000,
+ "train_loss": 0.2130824304008484
+ },
+ {
+ "epoch": 79,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.05569453909993172,
+ "mean_feedback_forward_cosine": 0.9991535387541118,
+ "mean_feedback_forward_relative_error": 0.040483501298647175,
+ "min_feedback_forward_cosine": 0.9984550476074219
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 30889,
+ "train_examples": 50000,
+ "train_loss": 0.21174772569656372
+ },
+ {
+ "epoch": 80,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.05345839634537697,
+ "mean_feedback_forward_cosine": 0.99921757296512,
+ "mean_feedback_forward_relative_error": 0.038921269343087546,
+ "min_feedback_forward_cosine": 0.9985762238502502
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 31280,
+ "train_examples": 50000,
+ "train_loss": 0.21362232236862183
+ },
+ {
+ "epoch": 81,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.05138672888278961,
+ "mean_feedback_forward_cosine": 0.9992752200678775,
+ "mean_feedback_forward_relative_error": 0.03746017117641474,
+ "min_feedback_forward_cosine": 0.9986842274665833
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 31671,
+ "train_examples": 50000,
+ "train_loss": 0.2059653507041931
+ },
+ {
+ "epoch": 82,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.04933715984225273,
+ "mean_feedback_forward_cosine": 0.9993301943728798,
+ "mean_feedback_forward_relative_error": 0.03601319264424475,
+ "min_feedback_forward_cosine": 0.9987869262695312
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 32062,
+ "train_examples": 50000,
+ "train_loss": 0.20739755942344665
+ },
+ {
+ "epoch": 83,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.04744468629360199,
+ "mean_feedback_forward_cosine": 0.9993793776160792,
+ "mean_feedback_forward_relative_error": 0.0346666251456267,
+ "min_feedback_forward_cosine": 0.9988775253295898
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 32453,
+ "train_examples": 50000,
+ "train_loss": 0.20578169294357299
+ },
+ {
+ "epoch": 84,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0454692542552948,
+ "mean_feedback_forward_cosine": 0.9994265280271831,
+ "mean_feedback_forward_relative_error": 0.03332244825402373,
+ "min_feedback_forward_cosine": 0.9989690184593201
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 32844,
+ "train_examples": 50000,
+ "train_loss": 0.21051717368125916
+ },
+ {
+ "epoch": 85,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0437321811914444,
+ "mean_feedback_forward_cosine": 0.9994687751719826,
+ "mean_feedback_forward_relative_error": 0.032071445578415146,
+ "min_feedback_forward_cosine": 0.9990459680557251
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 33235,
+ "train_examples": 50000,
+ "train_loss": 0.20646495123386382
+ },
+ {
+ "epoch": 86,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0419686958193779,
+ "mean_feedback_forward_cosine": 0.9995098427722329,
+ "mean_feedback_forward_relative_error": 0.030809350703891954,
+ "min_feedback_forward_cosine": 0.9991215467453003
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 33626,
+ "train_examples": 50000,
+ "train_loss": 0.2108780198287964
+ },
+ {
+ "epoch": 87,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.040337637066841125,
+ "mean_feedback_forward_cosine": 0.9995463333631817,
+ "mean_feedback_forward_relative_error": 0.02964061714316669,
+ "min_feedback_forward_cosine": 0.9991884231567383
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 34017,
+ "train_examples": 50000,
+ "train_loss": 0.20576658058166503
+ },
+ {
+ "epoch": 88,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0387629009783268,
+ "mean_feedback_forward_cosine": 0.999580044495432,
+ "mean_feedback_forward_relative_error": 0.02851661221173249,
+ "min_feedback_forward_cosine": 0.9992505311965942
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 34408,
+ "train_examples": 50000,
+ "train_loss": 0.20313345026493074
+ },
+ {
+ "epoch": 89,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03718485310673714,
+ "mean_feedback_forward_cosine": 0.9996134136852465,
+ "mean_feedback_forward_relative_error": 0.027364295417148816,
+ "min_feedback_forward_cosine": 0.9993102550506592
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 34799,
+ "train_examples": 50000,
+ "train_loss": 0.207878734331131
+ },
+ {
+ "epoch": 90,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03570469468832016,
+ "mean_feedback_forward_cosine": 0.999642412913473,
+ "mean_feedback_forward_relative_error": 0.026313925259991697,
+ "min_feedback_forward_cosine": 0.9993640780448914
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 35190,
+ "train_examples": 50000,
+ "train_loss": 0.20488694869041443
+ },
+ {
+ "epoch": 91,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034284189343452454,
+ "mean_feedback_forward_cosine": 0.9996700977024279,
+ "mean_feedback_forward_relative_error": 0.02527490321938929,
+ "min_feedback_forward_cosine": 0.9994134902954102
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 35581,
+ "train_examples": 50000,
+ "train_loss": 0.2048532602787018
+ },
+ {
+ "epoch": 92,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03291381523013115,
+ "mean_feedback_forward_cosine": 0.9996955394744873,
+ "mean_feedback_forward_relative_error": 0.024282213153415604,
+ "min_feedback_forward_cosine": 0.9994593262672424
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 35972,
+ "train_examples": 50000,
+ "train_loss": 0.20214042823314668
+ },
+ {
+ "epoch": 93,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.031641434878110886,
+ "mean_feedback_forward_cosine": 0.9997181735540691,
+ "mean_feedback_forward_relative_error": 0.023359945338023335,
+ "min_feedback_forward_cosine": 0.9995002150535583
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 36363,
+ "train_examples": 50000,
+ "train_loss": 0.20128610049247742
+ },
+ {
+ "epoch": 94,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.030388284474611282,
+ "mean_feedback_forward_cosine": 0.9997399794427972,
+ "mean_feedback_forward_relative_error": 0.022440489675653607,
+ "min_feedback_forward_cosine": 0.9995390176773071
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 36754,
+ "train_examples": 50000,
+ "train_loss": 0.20291523026943206
+ },
+ {
+ "epoch": 95,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.029230855405330658,
+ "mean_feedback_forward_cosine": 0.9997591533158955,
+ "mean_feedback_forward_relative_error": 0.021596452466359262,
+ "min_feedback_forward_cosine": 0.9995734691619873
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 37145,
+ "train_examples": 50000,
+ "train_loss": 0.20021228905200958
+ },
+ {
+ "epoch": 96,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.028075579553842545,
+ "mean_feedback_forward_cosine": 0.9997772103861758,
+ "mean_feedback_forward_relative_error": 0.02077175294490237,
+ "min_feedback_forward_cosine": 0.9996064901351929
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 37536,
+ "train_examples": 50000,
+ "train_loss": 0.20054742718696594
+ },
+ {
+ "epoch": 97,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02693236991763115,
+ "mean_feedback_forward_cosine": 0.9997943326046592,
+ "mean_feedback_forward_relative_error": 0.019957708832072586,
+ "min_feedback_forward_cosine": 0.9996378421783447
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 37927,
+ "train_examples": 50000,
+ "train_loss": 0.20103565855979919
+ },
+ {
+ "epoch": 98,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.025899013504385948,
+ "mean_feedback_forward_cosine": 0.9998096384500202,
+ "mean_feedback_forward_relative_error": 0.01920037510755815,
+ "min_feedback_forward_cosine": 0.9996651411056519
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 38318,
+ "train_examples": 50000,
+ "train_loss": 0.2001511568880081
+ },
+ {
+ "epoch": 99,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02488711290061474,
+ "mean_feedback_forward_cosine": 0.9998238494521693,
+ "mean_feedback_forward_relative_error": 0.018469309140192836,
+ "min_feedback_forward_cosine": 0.9996907711029053
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 38709,
+ "train_examples": 50000,
+ "train_loss": 0.19720856919288635
+ },
+ {
+ "epoch": 100,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023899318650364876,
+ "mean_feedback_forward_cosine": 0.999836874635596,
+ "mean_feedback_forward_relative_error": 0.017773398957950503,
+ "min_feedback_forward_cosine": 0.9997147917747498
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 39100,
+ "train_examples": 50000,
+ "train_loss": 0.19735110092163086
+ },
+ {
+ "epoch": 101,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02389109693467617,
+ "mean_feedback_forward_cosine": 0.9998370252157512,
+ "mean_feedback_forward_relative_error": 0.017763798174105193,
+ "min_feedback_forward_cosine": 0.9997150301933289
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 39491,
+ "train_examples": 50000,
+ "train_loss": 0.12442319479942322
+ },
+ {
+ "epoch": 102,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023886116221547127,
+ "mean_feedback_forward_cosine": 0.9998371005058289,
+ "mean_feedback_forward_relative_error": 0.01775621257624344,
+ "min_feedback_forward_cosine": 0.9997150897979736
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 39882,
+ "train_examples": 50000,
+ "train_loss": 0.09954114828824998
+ },
+ {
+ "epoch": 103,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0238815750926733,
+ "mean_feedback_forward_cosine": 0.9998371852071661,
+ "mean_feedback_forward_relative_error": 0.01774907479748914,
+ "min_feedback_forward_cosine": 0.9997152090072632
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 40273,
+ "train_examples": 50000,
+ "train_loss": 0.0886064899790287
+ },
+ {
+ "epoch": 104,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023877078667283058,
+ "mean_feedback_forward_cosine": 0.9998372542230707,
+ "mean_feedback_forward_relative_error": 0.017742220018255084,
+ "min_feedback_forward_cosine": 0.999715268611908
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 40664,
+ "train_examples": 50000,
+ "train_loss": 0.08099984602868557
+ },
+ {
+ "epoch": 105,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02387254498898983,
+ "mean_feedback_forward_cosine": 0.9998373702952736,
+ "mean_feedback_forward_relative_error": 0.01773549564868996,
+ "min_feedback_forward_cosine": 0.9997154474258423
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 41055,
+ "train_examples": 50000,
+ "train_loss": 0.07796452830791474
+ },
+ {
+ "epoch": 106,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023867983371019363,
+ "mean_feedback_forward_cosine": 0.9998374267628318,
+ "mean_feedback_forward_relative_error": 0.01772869549887745,
+ "min_feedback_forward_cosine": 0.9997155666351318
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 41446,
+ "train_examples": 50000,
+ "train_loss": 0.07303242535114288
+ },
+ {
+ "epoch": 107,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023863336071372032,
+ "mean_feedback_forward_cosine": 0.9998375522462946,
+ "mean_feedback_forward_relative_error": 0.0177221491834835,
+ "min_feedback_forward_cosine": 0.9997156858444214
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 41837,
+ "train_examples": 50000,
+ "train_loss": 0.0711615925681591
+ },
+ {
+ "epoch": 108,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023858843371272087,
+ "mean_feedback_forward_cosine": 0.9998375836171602,
+ "mean_feedback_forward_relative_error": 0.017715743301730407,
+ "min_feedback_forward_cosine": 0.9997157454490662
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 42228,
+ "train_examples": 50000,
+ "train_loss": 0.0673111107134819
+ },
+ {
+ "epoch": 109,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023854201659560204,
+ "mean_feedback_forward_cosine": 0.9998377059635363,
+ "mean_feedback_forward_relative_error": 0.01770933933163944,
+ "min_feedback_forward_cosine": 0.9997158646583557
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 42619,
+ "train_examples": 50000,
+ "train_loss": 0.06551601542472839
+ },
+ {
+ "epoch": 110,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02384946681559086,
+ "mean_feedback_forward_cosine": 0.999837759294008,
+ "mean_feedback_forward_relative_error": 0.01770310493578252,
+ "min_feedback_forward_cosine": 0.9997159838676453
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 43010,
+ "train_examples": 50000,
+ "train_loss": 0.06257788936972618
+ },
+ {
+ "epoch": 111,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023844989016652107,
+ "mean_feedback_forward_cosine": 0.9998378690920378,
+ "mean_feedback_forward_relative_error": 0.017696918454021215,
+ "min_feedback_forward_cosine": 0.9997161626815796
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 43401,
+ "train_examples": 50000,
+ "train_loss": 0.058294875597953794
+ },
+ {
+ "epoch": 112,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02384048141539097,
+ "mean_feedback_forward_cosine": 0.9998379443821154,
+ "mean_feedback_forward_relative_error": 0.017690715821165787,
+ "min_feedback_forward_cosine": 0.9997161626815796
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 43792,
+ "train_examples": 50000,
+ "train_loss": 0.05692799431443214
+ },
+ {
+ "epoch": 113,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02383587695658207,
+ "mean_feedback_forward_cosine": 0.9998380071238467,
+ "mean_feedback_forward_relative_error": 0.01768442557046288,
+ "min_feedback_forward_cosine": 0.9997162818908691
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 44183,
+ "train_examples": 50000,
+ "train_loss": 0.055019658786058426
+ },
+ {
+ "epoch": 114,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023831013590097427,
+ "mean_feedback_forward_cosine": 0.9998381294702229,
+ "mean_feedback_forward_relative_error": 0.01767808301864486,
+ "min_feedback_forward_cosine": 0.9997164607048035
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 44574,
+ "train_examples": 50000,
+ "train_loss": 0.05449852827012539
+ },
+ {
+ "epoch": 115,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02382613904774189,
+ "mean_feedback_forward_cosine": 0.9998382110344736,
+ "mean_feedback_forward_relative_error": 0.017671735491603613,
+ "min_feedback_forward_cosine": 0.9997165203094482
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 44965,
+ "train_examples": 50000,
+ "train_loss": 0.051638311725854875
+ },
+ {
+ "epoch": 116,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02382132038474083,
+ "mean_feedback_forward_cosine": 0.999838295735811,
+ "mean_feedback_forward_relative_error": 0.01766541619834147,
+ "min_feedback_forward_cosine": 0.9997166395187378
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 45356,
+ "train_examples": 50000,
+ "train_loss": 0.05075180400133133
+ },
+ {
+ "epoch": 117,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023816458880901337,
+ "mean_feedback_forward_cosine": 0.9998383710258886,
+ "mean_feedback_forward_relative_error": 0.017659144108428768,
+ "min_feedback_forward_cosine": 0.9997167587280273
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 45747,
+ "train_examples": 50000,
+ "train_loss": 0.04860522404909134
+ },
+ {
+ "epoch": 118,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02381158247590065,
+ "mean_feedback_forward_cosine": 0.9998384933722647,
+ "mean_feedback_forward_relative_error": 0.01765274339796681,
+ "min_feedback_forward_cosine": 0.9997168779373169
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 46138,
+ "train_examples": 50000,
+ "train_loss": 0.047121148210167885
+ },
+ {
+ "epoch": 119,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0238068625330925,
+ "mean_feedback_forward_cosine": 0.9998385592510826,
+ "mean_feedback_forward_relative_error": 0.017646590468326683,
+ "min_feedback_forward_cosine": 0.9997169375419617
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 46529,
+ "train_examples": 50000,
+ "train_loss": 0.04480014027833939
+ },
+ {
+ "epoch": 120,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023801730945706367,
+ "mean_feedback_forward_cosine": 0.9998386376782468,
+ "mean_feedback_forward_relative_error": 0.01764031075627396,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 46920,
+ "train_examples": 50000,
+ "train_loss": 0.04622192790746689
+ },
+ {
+ "epoch": 121,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0237965639680624,
+ "mean_feedback_forward_cosine": 0.9998387443391901,
+ "mean_feedback_forward_relative_error": 0.01763393043687469,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 47311,
+ "train_examples": 50000,
+ "train_loss": 0.044027461415529254
+ },
+ {
+ "epoch": 122,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023791490122675896,
+ "mean_feedback_forward_cosine": 0.9998388259034408,
+ "mean_feedback_forward_relative_error": 0.017627560754159565,
+ "min_feedback_forward_cosine": 0.9997173547744751
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 47702,
+ "train_examples": 50000,
+ "train_loss": 0.042354498464465144
+ },
+ {
+ "epoch": 123,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023786408826708794,
+ "mean_feedback_forward_cosine": 0.9998389262902109,
+ "mean_feedback_forward_relative_error": 0.01762126636152205,
+ "min_feedback_forward_cosine": 0.9997175335884094
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 48093,
+ "train_examples": 50000,
+ "train_loss": 0.04295653098106384
+ },
+ {
+ "epoch": 124,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023781437426805496,
+ "mean_feedback_forward_cosine": 0.9998390392253274,
+ "mean_feedback_forward_relative_error": 0.0176150009379183,
+ "min_feedback_forward_cosine": 0.999717652797699
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 48484,
+ "train_examples": 50000,
+ "train_loss": 0.04079782772898674
+ },
+ {
+ "epoch": 125,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023776011541485786,
+ "mean_feedback_forward_cosine": 0.9998391552975303,
+ "mean_feedback_forward_relative_error": 0.017608390900453453,
+ "min_feedback_forward_cosine": 0.9997177124023438
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 48875,
+ "train_examples": 50000,
+ "train_loss": 0.04124207681894303
+ },
+ {
+ "epoch": 126,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023770619183778763,
+ "mean_feedback_forward_cosine": 0.9998392054909154,
+ "mean_feedback_forward_relative_error": 0.017601775838748404,
+ "min_feedback_forward_cosine": 0.9997177124023438
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 49266,
+ "train_examples": 50000,
+ "train_loss": 0.03900240250706673
+ },
+ {
+ "epoch": 127,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023765409365296364,
+ "mean_feedback_forward_cosine": 0.999839324700205,
+ "mean_feedback_forward_relative_error": 0.017595374638116675,
+ "min_feedback_forward_cosine": 0.9997180700302124
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 49657,
+ "train_examples": 50000,
+ "train_loss": 0.037209171330928804
+ },
+ {
+ "epoch": 128,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023760031908750534,
+ "mean_feedback_forward_cosine": 0.9998394344982348,
+ "mean_feedback_forward_relative_error": 0.01758885564968774,
+ "min_feedback_forward_cosine": 0.9997181296348572
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 50048,
+ "train_examples": 50000,
+ "train_loss": 0.036044797257184985
+ },
+ {
+ "epoch": 129,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023754576221108437,
+ "mean_feedback_forward_cosine": 0.9998394846916199,
+ "mean_feedback_forward_relative_error": 0.017582232696249298,
+ "min_feedback_forward_cosine": 0.999718189239502
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 50439,
+ "train_examples": 50000,
+ "train_loss": 0.036427210804224015
+ },
+ {
+ "epoch": 130,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023748576641082764,
+ "mean_feedback_forward_cosine": 0.9998396258605154,
+ "mean_feedback_forward_relative_error": 0.017575340933705632,
+ "min_feedback_forward_cosine": 0.999718427658081
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 50830,
+ "train_examples": 50000,
+ "train_loss": 0.03773718276321888
+ },
+ {
+ "epoch": 131,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023743055760860443,
+ "mean_feedback_forward_cosine": 0.9998397105618527,
+ "mean_feedback_forward_relative_error": 0.017568696829441347,
+ "min_feedback_forward_cosine": 0.999718427658081
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 51221,
+ "train_examples": 50000,
+ "train_loss": 0.035393076547384265
+ },
+ {
+ "epoch": 132,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023737845942378044,
+ "mean_feedback_forward_cosine": 0.9998397827148438,
+ "mean_feedback_forward_relative_error": 0.017562087086078367,
+ "min_feedback_forward_cosine": 0.9997186660766602
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 51612,
+ "train_examples": 50000,
+ "train_loss": 0.033653816018104556
+ },
+ {
+ "epoch": 133,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02373233065009117,
+ "mean_feedback_forward_cosine": 0.9998399207466527,
+ "mean_feedback_forward_relative_error": 0.017555307547904943,
+ "min_feedback_forward_cosine": 0.9997189044952393
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 52003,
+ "train_examples": 50000,
+ "train_loss": 0.032840322127342225
+ },
+ {
+ "epoch": 134,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023726176470518112,
+ "mean_feedback_forward_cosine": 0.9998400399559423,
+ "mean_feedback_forward_relative_error": 0.01754812222268236,
+ "min_feedback_forward_cosine": 0.9997189044952393
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 52394,
+ "train_examples": 50000,
+ "train_loss": 0.035277710797786715
+ },
+ {
+ "epoch": 135,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023720096796751022,
+ "mean_feedback_forward_cosine": 0.999840143479799,
+ "mean_feedback_forward_relative_error": 0.017540947460618458,
+ "min_feedback_forward_cosine": 0.9997191429138184
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 52785,
+ "train_examples": 50000,
+ "train_loss": 0.03393282965540886
+ },
+ {
+ "epoch": 136,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023713892325758934,
+ "mean_feedback_forward_cosine": 0.9998402721003482,
+ "mean_feedback_forward_relative_error": 0.017533725372662668,
+ "min_feedback_forward_cosine": 0.9997192621231079
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 53176,
+ "train_examples": 50000,
+ "train_loss": 0.033962389070987704
+ },
+ {
+ "epoch": 137,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02370789274573326,
+ "mean_feedback_forward_cosine": 0.9998403787612915,
+ "mean_feedback_forward_relative_error": 0.01752667798121509,
+ "min_feedback_forward_cosine": 0.9997193813323975
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 53567,
+ "train_examples": 50000,
+ "train_loss": 0.03192068582713604
+ },
+ {
+ "epoch": 138,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023701651021838188,
+ "mean_feedback_forward_cosine": 0.999840469736802,
+ "mean_feedback_forward_relative_error": 0.01751929930852432,
+ "min_feedback_forward_cosine": 0.9997196197509766
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 53958,
+ "train_examples": 50000,
+ "train_loss": 0.032145514241456986
+ },
+ {
+ "epoch": 139,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023695718497037888,
+ "mean_feedback_forward_cosine": 0.9998405889460915,
+ "mean_feedback_forward_relative_error": 0.0175121195467287,
+ "min_feedback_forward_cosine": 0.9997196793556213
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 54349,
+ "train_examples": 50000,
+ "train_loss": 0.03121916231572628
+ },
+ {
+ "epoch": 140,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02369002439081669,
+ "mean_feedback_forward_cosine": 0.9998406956070348,
+ "mean_feedback_forward_relative_error": 0.017505143131864697,
+ "min_feedback_forward_cosine": 0.9997197389602661
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 54740,
+ "train_examples": 50000,
+ "train_loss": 0.029504041374325753
+ },
+ {
+ "epoch": 141,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023683952167630196,
+ "mean_feedback_forward_cosine": 0.9998408399130169,
+ "mean_feedback_forward_relative_error": 0.017497966139528313,
+ "min_feedback_forward_cosine": 0.99972003698349
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 55131,
+ "train_examples": 50000,
+ "train_loss": 0.02961511559844017
+ },
+ {
+ "epoch": 142,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023677857592701912,
+ "mean_feedback_forward_cosine": 0.9998409465739602,
+ "mean_feedback_forward_relative_error": 0.017490761648667485,
+ "min_feedback_forward_cosine": 0.9997200965881348
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 55522,
+ "train_examples": 50000,
+ "train_loss": 0.029603759601712226
+ },
+ {
+ "epoch": 143,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023671388626098633,
+ "mean_feedback_forward_cosine": 0.9998410563719901,
+ "mean_feedback_forward_relative_error": 0.017483263840212634,
+ "min_feedback_forward_cosine": 0.9997202754020691
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 55913,
+ "train_examples": 50000,
+ "train_loss": 0.029325441351532935
+ },
+ {
+ "epoch": 144,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023665640503168106,
+ "mean_feedback_forward_cosine": 0.9998411849925393,
+ "mean_feedback_forward_relative_error": 0.01747635232382699,
+ "min_feedback_forward_cosine": 0.9997204542160034
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 56304,
+ "train_examples": 50000,
+ "train_loss": 0.027311097153425216
+ },
+ {
+ "epoch": 145,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023659126833081245,
+ "mean_feedback_forward_cosine": 0.999841288516396,
+ "mean_feedback_forward_relative_error": 0.017468886596983986,
+ "min_feedback_forward_cosine": 0.9997206926345825
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 56695,
+ "train_examples": 50000,
+ "train_loss": 0.028823337376117707
+ },
+ {
+ "epoch": 146,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023652082309126854,
+ "mean_feedback_forward_cosine": 0.9998414234111184,
+ "mean_feedback_forward_relative_error": 0.01746097143347326,
+ "min_feedback_forward_cosine": 0.9997207522392273
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 57086,
+ "train_examples": 50000,
+ "train_loss": 0.028660762969851494
+ },
+ {
+ "epoch": 147,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023645399138331413,
+ "mean_feedback_forward_cosine": 0.9998415300720617,
+ "mean_feedback_forward_relative_error": 0.01745332255469341,
+ "min_feedback_forward_cosine": 0.9997209906578064
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 57477,
+ "train_examples": 50000,
+ "train_loss": 0.027866083926558493
+ },
+ {
+ "epoch": 148,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02363891899585724,
+ "mean_feedback_forward_cosine": 0.9998416681038705,
+ "mean_feedback_forward_relative_error": 0.017445795238018036,
+ "min_feedback_forward_cosine": 0.9997209310531616
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 57868,
+ "train_examples": 50000,
+ "train_loss": 0.027290484753251076
+ },
+ {
+ "epoch": 149,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023632267490029335,
+ "mean_feedback_forward_cosine": 0.9998417841760736,
+ "mean_feedback_forward_relative_error": 0.017438053202472235,
+ "min_feedback_forward_cosine": 0.9997212290763855
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 58259,
+ "train_examples": 50000,
+ "train_loss": 0.025752054628431796
+ },
+ {
+ "epoch": 150,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02362574264407158,
+ "mean_feedback_forward_cosine": 0.9998419033853632,
+ "mean_feedback_forward_relative_error": 0.01743022894094649,
+ "min_feedback_forward_cosine": 0.9997212886810303
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 58650,
+ "train_examples": 50000,
+ "train_loss": 0.025914649894833566
+ },
+ {
+ "epoch": 151,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02362567000091076,
+ "mean_feedback_forward_cosine": 0.9998418876999303,
+ "mean_feedback_forward_relative_error": 0.017430036598326343,
+ "min_feedback_forward_cosine": 0.9997212886810303
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 59041,
+ "train_examples": 50000,
+ "train_loss": 0.023274573529958726
+ },
+ {
+ "epoch": 152,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02362561784684658,
+ "mean_feedback_forward_cosine": 0.9998419096595362,
+ "mean_feedback_forward_relative_error": 0.017429864328158528,
+ "min_feedback_forward_cosine": 0.9997212290763855
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 59432,
+ "train_examples": 50000,
+ "train_loss": 0.021617045214772226
+ },
+ {
+ "epoch": 153,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023625561967492104,
+ "mean_feedback_forward_cosine": 0.9998418908370169,
+ "mean_feedback_forward_relative_error": 0.017429695146060305,
+ "min_feedback_forward_cosine": 0.9997212886810303
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 59823,
+ "train_examples": 50000,
+ "train_loss": 0.021565087901949883
+ },
+ {
+ "epoch": 154,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023625506088137627,
+ "mean_feedback_forward_cosine": 0.9998419127966228,
+ "mean_feedback_forward_relative_error": 0.017429523537621686,
+ "min_feedback_forward_cosine": 0.9997212886810303
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 60214,
+ "train_examples": 50000,
+ "train_loss": 0.021052006413936616
+ },
+ {
+ "epoch": 155,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023625463247299194,
+ "mean_feedback_forward_cosine": 0.9998419222078825,
+ "mean_feedback_forward_relative_error": 0.01742935232131889,
+ "min_feedback_forward_cosine": 0.9997212886810303
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 60605,
+ "train_examples": 50000,
+ "train_loss": 0.020187341000437738
+ },
+ {
+ "epoch": 156,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023625412955880165,
+ "mean_feedback_forward_cosine": 0.9998419347562288,
+ "mean_feedback_forward_relative_error": 0.017429182501999957,
+ "min_feedback_forward_cosine": 0.999721348285675
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 60996,
+ "train_examples": 50000,
+ "train_loss": 0.019765038191378117
+ },
+ {
+ "epoch": 157,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023625364527106285,
+ "mean_feedback_forward_cosine": 0.9998419378933153,
+ "mean_feedback_forward_relative_error": 0.0174290195205494,
+ "min_feedback_forward_cosine": 0.999721348285675
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 61387,
+ "train_examples": 50000,
+ "train_loss": 0.020545777665376665
+ },
+ {
+ "epoch": 158,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023625314235687256,
+ "mean_feedback_forward_cosine": 0.999841919070796,
+ "mean_feedback_forward_relative_error": 0.017428849235569175,
+ "min_feedback_forward_cosine": 0.9997212886810303
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 61778,
+ "train_examples": 50000,
+ "train_loss": 0.019909772011041642
+ },
+ {
+ "epoch": 159,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023625262081623077,
+ "mean_feedback_forward_cosine": 0.9998419222078825,
+ "mean_feedback_forward_relative_error": 0.017428683239574496,
+ "min_feedback_forward_cosine": 0.999721348285675
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 62169,
+ "train_examples": 50000,
+ "train_loss": 0.019450022747516633
+ },
+ {
+ "epoch": 160,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023625215515494347,
+ "mean_feedback_forward_cosine": 0.9998419065224496,
+ "mean_feedback_forward_relative_error": 0.01742852018459847,
+ "min_feedback_forward_cosine": 0.9997214078903198
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 62560,
+ "train_examples": 50000,
+ "train_loss": 0.018635217373669147
+ },
+ {
+ "epoch": 161,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023625168949365616,
+ "mean_feedback_forward_cosine": 0.9998419284820557,
+ "mean_feedback_forward_relative_error": 0.01742836176172683,
+ "min_feedback_forward_cosine": 0.999721348285675
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 62951,
+ "train_examples": 50000,
+ "train_loss": 0.019771754233539104
+ },
+ {
+ "epoch": 162,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023625120520591736,
+ "mean_feedback_forward_cosine": 0.9998419033853632,
+ "mean_feedback_forward_relative_error": 0.017428197089190547,
+ "min_feedback_forward_cosine": 0.9997212886810303
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 63342,
+ "train_examples": 50000,
+ "train_loss": 0.019207457506358622
+ },
+ {
+ "epoch": 163,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023625075817108154,
+ "mean_feedback_forward_cosine": 0.9998419284820557,
+ "mean_feedback_forward_relative_error": 0.017428037220318066,
+ "min_feedback_forward_cosine": 0.9997214078903198
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 63733,
+ "train_examples": 50000,
+ "train_loss": 0.01966070913493633
+ },
+ {
+ "epoch": 164,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023625025525689125,
+ "mean_feedback_forward_cosine": 0.9998419159337094,
+ "mean_feedback_forward_relative_error": 0.017427871469408274,
+ "min_feedback_forward_cosine": 0.9997212886810303
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 64124,
+ "train_examples": 50000,
+ "train_loss": 0.018938596695363522
+ },
+ {
+ "epoch": 165,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023624984547495842,
+ "mean_feedback_forward_cosine": 0.9998419253449691,
+ "mean_feedback_forward_relative_error": 0.017427710179043442,
+ "min_feedback_forward_cosine": 0.9997212886810303
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 64515,
+ "train_examples": 50000,
+ "train_loss": 0.01795565570920706
+ },
+ {
+ "epoch": 166,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02362494170665741,
+ "mean_feedback_forward_cosine": 0.9998419284820557,
+ "mean_feedback_forward_relative_error": 0.017427550702306786,
+ "min_feedback_forward_cosine": 0.999721348285675
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 64906,
+ "train_examples": 50000,
+ "train_loss": 0.01870345171034336
+ },
+ {
+ "epoch": 167,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02362489514052868,
+ "mean_feedback_forward_cosine": 0.9998419441674885,
+ "mean_feedback_forward_relative_error": 0.0174273925980455,
+ "min_feedback_forward_cosine": 0.9997212886810303
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 65297,
+ "train_examples": 50000,
+ "train_loss": 0.017115880472362042
+ },
+ {
+ "epoch": 168,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023624848574399948,
+ "mean_feedback_forward_cosine": 0.9998419378933153,
+ "mean_feedback_forward_relative_error": 0.017427234836903057,
+ "min_feedback_forward_cosine": 0.9997214078903198
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 65688,
+ "train_examples": 50000,
+ "train_loss": 0.01920094470024109
+ },
+ {
+ "epoch": 169,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023624802008271217,
+ "mean_feedback_forward_cosine": 0.9998419410304019,
+ "mean_feedback_forward_relative_error": 0.01742707433080987,
+ "min_feedback_forward_cosine": 0.9997214078903198
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 66079,
+ "train_examples": 50000,
+ "train_loss": 0.01911835408836603
+ },
+ {
+ "epoch": 170,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023624759167432785,
+ "mean_feedback_forward_cosine": 0.9998419504416617,
+ "mean_feedback_forward_relative_error": 0.017426916765735337,
+ "min_feedback_forward_cosine": 0.9997212886810303
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 66470,
+ "train_examples": 50000,
+ "train_loss": 0.01843534249663353
+ },
+ {
+ "epoch": 171,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023624710738658905,
+ "mean_feedback_forward_cosine": 0.9998419316191423,
+ "mean_feedback_forward_relative_error": 0.01742676098978049,
+ "min_feedback_forward_cosine": 0.9997212886810303
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 66861,
+ "train_examples": 50000,
+ "train_loss": 0.01794396436005831
+ },
+ {
+ "epoch": 172,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02362467721104622,
+ "mean_feedback_forward_cosine": 0.9998419598529213,
+ "mean_feedback_forward_relative_error": 0.01742659862104215,
+ "min_feedback_forward_cosine": 0.999721348285675
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 67252,
+ "train_examples": 50000,
+ "train_loss": 0.017799743198156356
+ },
+ {
+ "epoch": 173,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023624630644917488,
+ "mean_feedback_forward_cosine": 0.9998419410304019,
+ "mean_feedback_forward_relative_error": 0.01742643122806361,
+ "min_feedback_forward_cosine": 0.9997214078903198
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 67643,
+ "train_examples": 50000,
+ "train_loss": 0.016940128788501024
+ },
+ {
+ "epoch": 174,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023624585941433907,
+ "mean_feedback_forward_cosine": 0.9998419347562288,
+ "mean_feedback_forward_relative_error": 0.01742627199641184,
+ "min_feedback_forward_cosine": 0.9997212886810303
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 68034,
+ "train_examples": 50000,
+ "train_loss": 0.017658958861529826
+ },
+ {
+ "epoch": 175,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023624535650014877,
+ "mean_feedback_forward_cosine": 0.9998419347562288,
+ "mean_feedback_forward_relative_error": 0.017426116563575834,
+ "min_feedback_forward_cosine": 0.9997214078903198
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 68425,
+ "train_examples": 50000,
+ "train_loss": 0.01826230070203543
+ },
+ {
+ "epoch": 176,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023624492809176445,
+ "mean_feedback_forward_cosine": 0.9998419755383542,
+ "mean_feedback_forward_relative_error": 0.017425962405181246,
+ "min_feedback_forward_cosine": 0.9997212886810303
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 68816,
+ "train_examples": 50000,
+ "train_loss": 0.018668894878029824
+ },
+ {
+ "epoch": 177,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023624451830983162,
+ "mean_feedback_forward_cosine": 0.9998419410304019,
+ "mean_feedback_forward_relative_error": 0.017425803394105873,
+ "min_feedback_forward_cosine": 0.9997214078903198
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 69207,
+ "train_examples": 50000,
+ "train_loss": 0.01702467140108347
+ },
+ {
+ "epoch": 178,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02362440526485443,
+ "mean_feedback_forward_cosine": 0.9998419347562288,
+ "mean_feedback_forward_relative_error": 0.017425649039643377,
+ "min_feedback_forward_cosine": 0.9997214078903198
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 69598,
+ "train_examples": 50000,
+ "train_loss": 0.016476852412819862
+ },
+ {
+ "epoch": 179,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02362436056137085,
+ "mean_feedback_forward_cosine": 0.9998419504416617,
+ "mean_feedback_forward_relative_error": 0.01742549343524795,
+ "min_feedback_forward_cosine": 0.9997214078903198
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 69989,
+ "train_examples": 50000,
+ "train_loss": 0.016713480235040188
+ },
+ {
+ "epoch": 180,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02362431213259697,
+ "mean_feedback_forward_cosine": 0.9998419661270944,
+ "mean_feedback_forward_relative_error": 0.01742533601722435,
+ "min_feedback_forward_cosine": 0.9997214078903198
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 70380,
+ "train_examples": 50000,
+ "train_loss": 0.017687734822034836
+ },
+ {
+ "epoch": 181,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023624269291758537,
+ "mean_feedback_forward_cosine": 0.9998419441674885,
+ "mean_feedback_forward_relative_error": 0.017425186294866234,
+ "min_feedback_forward_cosine": 0.9997214078903198
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 70771,
+ "train_examples": 50000,
+ "train_loss": 0.017140326766371727
+ },
+ {
+ "epoch": 182,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023624228313565254,
+ "mean_feedback_forward_cosine": 0.9998419410304019,
+ "mean_feedback_forward_relative_error": 0.017425029489554857,
+ "min_feedback_forward_cosine": 0.9997214078903198
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 71162,
+ "train_examples": 50000,
+ "train_loss": 0.017491330931782723
+ },
+ {
+ "epoch": 183,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023624183610081673,
+ "mean_feedback_forward_cosine": 0.9998419410304019,
+ "mean_feedback_forward_relative_error": 0.01742487182644637,
+ "min_feedback_forward_cosine": 0.9997214078903198
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 71553,
+ "train_examples": 50000,
+ "train_loss": 0.016248736631572246
+ },
+ {
+ "epoch": 184,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02362413890659809,
+ "mean_feedback_forward_cosine": 0.9998419410304019,
+ "mean_feedback_forward_relative_error": 0.0174247119330654,
+ "min_feedback_forward_cosine": 0.9997212886810303
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 71944,
+ "train_examples": 50000,
+ "train_loss": 0.01694776742488146
+ },
+ {
+ "epoch": 185,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023624099791049957,
+ "mean_feedback_forward_cosine": 0.999841969264181,
+ "mean_feedback_forward_relative_error": 0.017424556181619043,
+ "min_feedback_forward_cosine": 0.9997212886810303
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 72335,
+ "train_examples": 50000,
+ "train_loss": 0.01671396476060152
+ },
+ {
+ "epoch": 186,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023624062538146973,
+ "mean_feedback_forward_cosine": 0.9998419535787482,
+ "mean_feedback_forward_relative_error": 0.017424403224140406,
+ "min_feedback_forward_cosine": 0.999721348285675
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 72726,
+ "train_examples": 50000,
+ "train_loss": 0.016440652879476547
+ },
+ {
+ "epoch": 187,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02362402155995369,
+ "mean_feedback_forward_cosine": 0.9998419535787482,
+ "mean_feedback_forward_relative_error": 0.017424247521711022,
+ "min_feedback_forward_cosine": 0.9997214078903198
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 73117,
+ "train_examples": 50000,
+ "train_loss": 0.01723799763441086
+ },
+ {
+ "epoch": 188,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023623984307050705,
+ "mean_feedback_forward_cosine": 0.9998419504416617,
+ "mean_feedback_forward_relative_error": 0.017424092505519326,
+ "min_feedback_forward_cosine": 0.9997212886810303
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 73508,
+ "train_examples": 50000,
+ "train_loss": 0.01625235303938389
+ },
+ {
+ "epoch": 189,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023623935878276825,
+ "mean_feedback_forward_cosine": 0.9998419849496138,
+ "mean_feedback_forward_relative_error": 0.01742394516048463,
+ "min_feedback_forward_cosine": 0.9997214078903198
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 73899,
+ "train_examples": 50000,
+ "train_loss": 0.017007624327540397
+ },
+ {
+ "epoch": 190,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023623894900083542,
+ "mean_feedback_forward_cosine": 0.9998419567158348,
+ "mean_feedback_forward_relative_error": 0.017423789899208043,
+ "min_feedback_forward_cosine": 0.9997214078903198
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 74290,
+ "train_examples": 50000,
+ "train_loss": 0.01589294947952032
+ },
+ {
+ "epoch": 191,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023623844608664513,
+ "mean_feedback_forward_cosine": 0.9998419724012676,
+ "mean_feedback_forward_relative_error": 0.017423640127832953,
+ "min_feedback_forward_cosine": 0.9997214078903198
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 74681,
+ "train_examples": 50000,
+ "train_loss": 0.017078018230199814
+ },
+ {
+ "epoch": 192,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02362380363047123,
+ "mean_feedback_forward_cosine": 0.9998419598529213,
+ "mean_feedback_forward_relative_error": 0.01742348793011747,
+ "min_feedback_forward_cosine": 0.9997212886810303
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 75072,
+ "train_examples": 50000,
+ "train_loss": 0.017073209224045278
+ },
+ {
+ "epoch": 193,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023623768240213394,
+ "mean_feedback_forward_cosine": 0.9998419567158348,
+ "mean_feedback_forward_relative_error": 0.017423336369622695,
+ "min_feedback_forward_cosine": 0.9997214078903198
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 75463,
+ "train_examples": 50000,
+ "train_loss": 0.015712709036171436
+ },
+ {
+ "epoch": 194,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023623721674084663,
+ "mean_feedback_forward_cosine": 0.9998419441674885,
+ "mean_feedback_forward_relative_error": 0.017423190348046392,
+ "min_feedback_forward_cosine": 0.999721348285675
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 75854,
+ "train_examples": 50000,
+ "train_loss": 0.0167135486972332
+ },
+ {
+ "epoch": 195,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02362368255853653,
+ "mean_feedback_forward_cosine": 0.9998419880867004,
+ "mean_feedback_forward_relative_error": 0.017423034057413276,
+ "min_feedback_forward_cosine": 0.9997214078903198
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 76245,
+ "train_examples": 50000,
+ "train_loss": 0.016926228197813035
+ },
+ {
+ "epoch": 196,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023623641580343246,
+ "mean_feedback_forward_cosine": 0.9998419786754408,
+ "mean_feedback_forward_relative_error": 0.017422879776476247,
+ "min_feedback_forward_cosine": 0.9997214674949646
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 76636,
+ "train_examples": 50000,
+ "train_loss": 0.015681275087296963
+ },
+ {
+ "epoch": 197,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02362360619008541,
+ "mean_feedback_forward_cosine": 0.9998419316191423,
+ "mean_feedback_forward_relative_error": 0.017422724368148727,
+ "min_feedback_forward_cosine": 0.999721348285675
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 77027,
+ "train_examples": 50000,
+ "train_loss": 0.0157519710201025
+ },
+ {
+ "epoch": 198,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023623565211892128,
+ "mean_feedback_forward_cosine": 0.9998419598529213,
+ "mean_feedback_forward_relative_error": 0.017422574400705725,
+ "min_feedback_forward_cosine": 0.999721348285675
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 77418,
+ "train_examples": 50000,
+ "train_loss": 0.015686207499504088
+ },
+ {
+ "epoch": 199,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023623526096343994,
+ "mean_feedback_forward_cosine": 0.9998419818125273,
+ "mean_feedback_forward_relative_error": 0.01742242078149789,
+ "min_feedback_forward_cosine": 0.9997214078903198
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 77809,
+ "train_examples": 50000,
+ "train_loss": 0.015580578749477863
+ },
+ {
+ "epoch": 200,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02362348884344101,
+ "mean_feedback_forward_cosine": 0.9998419629900079,
+ "mean_feedback_forward_relative_error": 0.017422264711441177,
+ "min_feedback_forward_cosine": 0.9997214078903198
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 78200,
+ "train_examples": 50000,
+ "train_loss": 0.015610226914286613
+ }
+ ],
+ "evaluation_protocol": {
+ "test_evaluations": 1,
+ "test_used_for_selection": false,
+ "validation_evaluations": 0
+ },
+ "final": {
+ "accuracy": 0.9087,
+ "epoch": 200,
+ "evaluation_split": "test",
+ "finite": true,
+ "loss": 0.4213775695800781,
+ "step": 78200
+ },
+ "hardware": {
+ "cuda_device_name": "NVIDIA GeForce GTX 1080",
+ "cuda_visible_devices": "5",
+ "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": 12,
+ "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": 0.7231495380401611,
+ "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": 6212.18221783638,
+ "train_wall_s": 6210.074757575989
+ },
+ "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
+ }
+}