summaryrefslogtreecommitdiff
path: root/results/kp_dynamic_projection_confirmation/seed10_clean_kp.json
diff options
context:
space:
mode:
Diffstat (limited to 'results/kp_dynamic_projection_confirmation/seed10_clean_kp.json')
-rw-r--r--results/kp_dynamic_projection_confirmation/seed10_clean_kp.json3258
1 files changed, 3258 insertions, 0 deletions
diff --git a/results/kp_dynamic_projection_confirmation/seed10_clean_kp.json b/results/kp_dynamic_projection_confirmation/seed10_clean_kp.json
new file mode 100644
index 0000000..0dafe3b
--- /dev/null
+++ b/results/kp_dynamic_projection_confirmation/seed10_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": 10,
+ "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/seed10_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": 10,
+ "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": 0.9996943175792694,
+ "feedback_forward_cosine": [
+ 0.9998425245285034,
+ 0.9998425245285034,
+ 0.9998345971107483,
+ 0.9998288154602051,
+ 0.9998345971107483,
+ 0.9997845888137817,
+ 0.9998021721839905,
+ 0.9998700618743896,
+ 0.9998346567153931,
+ 0.9997909069061279,
+ 0.9998584985733032,
+ 0.999813437461853,
+ 0.9998350143432617,
+ 0.999904453754425,
+ 0.9998966455459595,
+ 0.9998763799667358,
+ 0.9998807907104492,
+ 0.9997172951698303,
+ 0.999988317489624
+ ],
+ "feedback_forward_norm_ratio": [
+ 0.9995854496955872,
+ 1.0009312629699707,
+ 0.9996626973152161,
+ 1.0009467601776123,
+ 0.9994809627532959,
+ 1.000537395477295,
+ 0.9993156790733337,
+ 1.0004427433013916,
+ 0.9995880126953125,
+ 1.0010364055633545,
+ 0.9996651411056519,
+ 1.0011072158813477,
+ 0.9993873834609985,
+ 1.0000609159469604,
+ 0.9999977350234985,
+ 1.0003987550735474,
+ 1.0000828504562378,
+ 1.0007964372634888,
+ 1.0027433633804321
+ ],
+ "feedback_forward_relative_error": [
+ 0.01775248721241951,
+ 0.017780128866434097,
+ 0.018192417919635773,
+ 0.0185343399643898,
+ 0.018192430958151817,
+ 0.02076704613864422,
+ 0.019896846264600754,
+ 0.016129642724990845,
+ 0.018187053501605988,
+ 0.020488634705543518,
+ 0.01682351343333721,
+ 0.019359398633241653,
+ 0.018172236159443855,
+ 0.013829465955495834,
+ 0.01438230462372303,
+ 0.015732765197753906,
+ 0.015434577129781246,
+ 0.02380364201962948,
+ 0.005563169717788696
+ ],
+ "innovation_negative_gradient_cosine": [
+ 0.999703049659729,
+ 0.9996821880340576,
+ 0.9996962547302246,
+ 0.9996733069419861,
+ 0.9997035264968872,
+ 0.9997075796127319,
+ 0.9997260570526123,
+ 0.9997570514678955,
+ 0.9997557401657104,
+ 0.9997051358222961,
+ 0.99976646900177,
+ 0.9997504353523254,
+ 0.9997883439064026,
+ 0.9998454451560974,
+ 0.9998633861541748,
+ 0.9998831152915955,
+ 0.9999251365661621,
+ 0.9999701976776123,
+ 0.8750146627426147
+ ],
+ "mean_feedback_forward_cosine": 0.9998440146446228,
+ "mean_feedback_forward_relative_error": 0.017316952690874274,
+ "normalization_state": "training_batch_stats_without_running_update",
+ "raw_negative_gradient_cosine": [
+ 0.999703049659729,
+ 0.9996821880340576,
+ 0.9996962547302246,
+ 0.9996733069419861,
+ 0.9997035264968872,
+ 0.9997075796127319,
+ 0.9997260570526123,
+ 0.9997570514678955,
+ 0.9997557401657104,
+ 0.9997051358222961,
+ 0.99976646900177,
+ 0.9997504353523254,
+ 0.9997883439064026,
+ 0.9998454451560974,
+ 0.9998633861541748,
+ 0.9998831152915955,
+ 0.9999251365661621,
+ 0.9999701976776123,
+ 0.8750146627426147
+ ],
+ "teaching_negative_gradient_cosine": [
+ 0.999703049659729,
+ 0.9996821880340576,
+ 0.9996962547302246,
+ 0.9996733069419861,
+ 0.9997035264968872,
+ 0.9997075796127319,
+ 0.9997260570526123,
+ 0.9997570514678955,
+ 0.9997557401657104,
+ 0.9997051358222961,
+ 0.99976646900177,
+ 0.9997504353523254,
+ 0.9997883439064026,
+ 0.9998454451560974,
+ 0.9998633861541748,
+ 0.9998831152915955,
+ 0.9999251365661621,
+ 0.9999701976776123,
+ 0.8750146627426147
+ ],
+ "wall_s": 0.09648609161376953
+ },
+ "epochs": [
+ {
+ "epoch": 1,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.3041185140609741,
+ "mean_feedback_forward_cosine": 0.2736612542679435,
+ "mean_feedback_forward_relative_error": 1.2024953898630644,
+ "min_feedback_forward_cosine": 0.15773168206214905
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 391,
+ "train_examples": 50000,
+ "train_loss": 1.8356010372543334
+ },
+ {
+ "epoch": 2,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.243371844291687,
+ "mean_feedback_forward_cosine": 0.4222466114320253,
+ "mean_feedback_forward_relative_error": 1.0754974955006649,
+ "min_feedback_forward_cosine": 0.24570325016975403
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 782,
+ "train_examples": 50000,
+ "train_loss": 1.2715008509635926
+ },
+ {
+ "epoch": 3,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.1895959377288818,
+ "mean_feedback_forward_cosine": 0.5245403443512163,
+ "mean_feedback_forward_relative_error": 0.9786683587651503,
+ "min_feedback_forward_cosine": 0.3132213354110718
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 1173,
+ "train_examples": 50000,
+ "train_loss": 0.9838771463012695
+ },
+ {
+ "epoch": 4,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.141883373260498,
+ "mean_feedback_forward_cosine": 0.5945022388508445,
+ "mean_feedback_forward_relative_error": 0.9046765205107237,
+ "min_feedback_forward_cosine": 0.3691554665565491
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 1564,
+ "train_examples": 50000,
+ "train_loss": 0.8258294191360473
+ },
+ {
+ "epoch": 5,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.1009702682495117,
+ "mean_feedback_forward_cosine": 0.6463952660560608,
+ "mean_feedback_forward_relative_error": 0.8451103222997565,
+ "min_feedback_forward_cosine": 0.41622984409332275
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 1955,
+ "train_examples": 50000,
+ "train_loss": 0.7246334367179871
+ },
+ {
+ "epoch": 6,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.059396505355835,
+ "mean_feedback_forward_cosine": 0.6886242486928639,
+ "mean_feedback_forward_relative_error": 0.7933045343348855,
+ "min_feedback_forward_cosine": 0.46115246415138245
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 2346,
+ "train_examples": 50000,
+ "train_loss": 0.6571639463996887
+ },
+ {
+ "epoch": 7,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.0216901302337646,
+ "mean_feedback_forward_cosine": 0.7222898382889597,
+ "mean_feedback_forward_relative_error": 0.7487508936932212,
+ "min_feedback_forward_cosine": 0.49993962049484253
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 2737,
+ "train_examples": 50000,
+ "train_loss": 0.6064288543891907
+ },
+ {
+ "epoch": 8,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.9850110411643982,
+ "mean_feedback_forward_cosine": 0.7514332783849615,
+ "mean_feedback_forward_relative_error": 0.7081342119919626,
+ "min_feedback_forward_cosine": 0.5357208251953125
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 3128,
+ "train_examples": 50000,
+ "train_loss": 0.574867534160614
+ },
+ {
+ "epoch": 9,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.9485263228416443,
+ "mean_feedback_forward_cosine": 0.7760953903198242,
+ "mean_feedback_forward_relative_error": 0.6718429358381974,
+ "min_feedback_forward_cosine": 0.5699862241744995
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 3519,
+ "train_examples": 50000,
+ "train_loss": 0.5442277415084839
+ },
+ {
+ "epoch": 10,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.9140889644622803,
+ "mean_feedback_forward_cosine": 0.7968643213573255,
+ "mean_feedback_forward_relative_error": 0.6396513330309015,
+ "min_feedback_forward_cosine": 0.601062536239624
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 3910,
+ "train_examples": 50000,
+ "train_loss": 0.5144776938438416
+ },
+ {
+ "epoch": 11,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.8813732862472534,
+ "mean_feedback_forward_cosine": 0.8151358616979498,
+ "mean_feedback_forward_relative_error": 0.6098568298314747,
+ "min_feedback_forward_cosine": 0.6292847990989685
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 4301,
+ "train_examples": 50000,
+ "train_loss": 0.4899724453735352
+ },
+ {
+ "epoch": 12,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.848876416683197,
+ "mean_feedback_forward_cosine": 0.831631180487181,
+ "mean_feedback_forward_relative_error": 0.5815863091694681,
+ "min_feedback_forward_cosine": 0.65630042552948
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 4692,
+ "train_examples": 50000,
+ "train_loss": 0.4721221300506592
+ },
+ {
+ "epoch": 13,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.8168927431106567,
+ "mean_feedback_forward_cosine": 0.8462759601442438,
+ "mean_feedback_forward_relative_error": 0.5554791563435605,
+ "min_feedback_forward_cosine": 0.6820494532585144
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 5083,
+ "train_examples": 50000,
+ "train_loss": 0.45774565432548525
+ },
+ {
+ "epoch": 14,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.785374641418457,
+ "mean_feedback_forward_cosine": 0.8596692806796024,
+ "mean_feedback_forward_relative_error": 0.5303741740553003,
+ "min_feedback_forward_cosine": 0.7061700820922852
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 5474,
+ "train_examples": 50000,
+ "train_loss": 0.4391380731391907
+ },
+ {
+ "epoch": 15,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.7563701272010803,
+ "mean_feedback_forward_cosine": 0.8713308698252628,
+ "mean_feedback_forward_relative_error": 0.5075985993209639,
+ "min_feedback_forward_cosine": 0.7274447679519653
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 5865,
+ "train_examples": 50000,
+ "train_loss": 0.4230655994129181
+ },
+ {
+ "epoch": 16,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.7266731858253479,
+ "mean_feedback_forward_cosine": 0.8819958097056338,
+ "mean_feedback_forward_relative_error": 0.4858224242925644,
+ "min_feedback_forward_cosine": 0.7484070658683777
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 6256,
+ "train_examples": 50000,
+ "train_loss": 0.4118123607349396
+ },
+ {
+ "epoch": 17,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.6991254091262817,
+ "mean_feedback_forward_cosine": 0.891421914100647,
+ "mean_feedback_forward_relative_error": 0.4657441009032099,
+ "min_feedback_forward_cosine": 0.7670483589172363
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 6647,
+ "train_examples": 50000,
+ "train_loss": 0.4013237077999115
+ },
+ {
+ "epoch": 18,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.6734879016876221,
+ "mean_feedback_forward_cosine": 0.9000662973052577,
+ "mean_feedback_forward_relative_error": 0.44667904941659226,
+ "min_feedback_forward_cosine": 0.7839051485061646
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 7038,
+ "train_examples": 50000,
+ "train_loss": 0.3887851303672791
+ },
+ {
+ "epoch": 19,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.6473090052604675,
+ "mean_feedback_forward_cosine": 0.9079619457847193,
+ "mean_feedback_forward_relative_error": 0.4283676633709355,
+ "min_feedback_forward_cosine": 0.8000421524047852
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 7429,
+ "train_examples": 50000,
+ "train_loss": 0.38309918633461
+ },
+ {
+ "epoch": 20,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.6222605109214783,
+ "mean_feedback_forward_cosine": 0.9152506746743855,
+ "mean_feedback_forward_relative_error": 0.41083644330501556,
+ "min_feedback_forward_cosine": 0.8149977922439575
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 7820,
+ "train_examples": 50000,
+ "train_loss": 0.37596964239120484
+ },
+ {
+ "epoch": 21,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.5985783934593201,
+ "mean_feedback_forward_cosine": 0.9218225102675589,
+ "mean_feedback_forward_relative_error": 0.3943716177814885,
+ "min_feedback_forward_cosine": 0.8287612199783325
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 8211,
+ "train_examples": 50000,
+ "train_loss": 0.3586363932991028
+ },
+ {
+ "epoch": 22,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.576032817363739,
+ "mean_feedback_forward_cosine": 0.9277743634424711,
+ "mean_feedback_forward_relative_error": 0.3788731788334094,
+ "min_feedback_forward_cosine": 0.8413866758346558
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 8602,
+ "train_examples": 50000,
+ "train_loss": 0.35420421438217164
+ },
+ {
+ "epoch": 23,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.5534531474113464,
+ "mean_feedback_forward_cosine": 0.9333119988441467,
+ "mean_feedback_forward_relative_error": 0.3638961550436522,
+ "min_feedback_forward_cosine": 0.853378415107727
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 8993,
+ "train_examples": 50000,
+ "train_loss": 0.34655102869033816
+ },
+ {
+ "epoch": 24,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.5311400294303894,
+ "mean_feedback_forward_cosine": 0.9385892152786255,
+ "mean_feedback_forward_relative_error": 0.3490081515751387,
+ "min_feedback_forward_cosine": 0.8648223876953125
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 9384,
+ "train_examples": 50000,
+ "train_loss": 0.34269681767463683
+ },
+ {
+ "epoch": 25,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.510734498500824,
+ "mean_feedback_forward_cosine": 0.9431674480438232,
+ "mean_feedback_forward_relative_error": 0.33550663370835154,
+ "min_feedback_forward_cosine": 0.8749117851257324
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 9775,
+ "train_examples": 50000,
+ "train_loss": 0.33644331916809084
+ },
+ {
+ "epoch": 26,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.4906480610370636,
+ "mean_feedback_forward_cosine": 0.9474997238108986,
+ "mean_feedback_forward_relative_error": 0.32233510676183197,
+ "min_feedback_forward_cosine": 0.8844285011291504
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 10166,
+ "train_examples": 50000,
+ "train_loss": 0.3293344172334671
+ },
+ {
+ "epoch": 27,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.4712061882019043,
+ "mean_feedback_forward_cosine": 0.9515334179526881,
+ "mean_feedback_forward_relative_error": 0.30956344149614634,
+ "min_feedback_forward_cosine": 0.8932749629020691
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 10557,
+ "train_examples": 50000,
+ "train_loss": 0.32324320629119874
+ },
+ {
+ "epoch": 28,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.45228156447410583,
+ "mean_feedback_forward_cosine": 0.9551924969020643,
+ "mean_feedback_forward_relative_error": 0.2975124677545146,
+ "min_feedback_forward_cosine": 0.9016437530517578
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 10948,
+ "train_examples": 50000,
+ "train_loss": 0.3145335978984833
+ },
+ {
+ "epoch": 29,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.43441838026046753,
+ "mean_feedback_forward_cosine": 0.9585838223758497,
+ "mean_feedback_forward_relative_error": 0.28590018733551625,
+ "min_feedback_forward_cosine": 0.9091375470161438
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 11339,
+ "train_examples": 50000,
+ "train_loss": 0.31292813460826874
+ },
+ {
+ "epoch": 30,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.416961133480072,
+ "mean_feedback_forward_cosine": 0.9617287202885276,
+ "mean_feedback_forward_relative_error": 0.2747007573121472,
+ "min_feedback_forward_cosine": 0.9161843061447144
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 11730,
+ "train_examples": 50000,
+ "train_loss": 0.30754787801742556
+ },
+ {
+ "epoch": 31,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.39982619881629944,
+ "mean_feedback_forward_cosine": 0.9646200694535908,
+ "mean_feedback_forward_relative_error": 0.26402254912414047,
+ "min_feedback_forward_cosine": 0.9228807687759399
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 12121,
+ "train_examples": 50000,
+ "train_loss": 0.3073337629413605
+ },
+ {
+ "epoch": 32,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.3841182291507721,
+ "mean_feedback_forward_cosine": 0.9672255829760903,
+ "mean_feedback_forward_relative_error": 0.25402774348070745,
+ "min_feedback_forward_cosine": 0.928751528263092
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 12512,
+ "train_examples": 50000,
+ "train_loss": 0.2970037853050232
+ },
+ {
+ "epoch": 33,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.36938709020614624,
+ "mean_feedback_forward_cosine": 0.9696529570378756,
+ "mean_feedback_forward_relative_error": 0.24437143928126284,
+ "min_feedback_forward_cosine": 0.9339888095855713
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 12903,
+ "train_examples": 50000,
+ "train_loss": 0.29162209347724916
+ },
+ {
+ "epoch": 34,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.3549087941646576,
+ "mean_feedback_forward_cosine": 0.9719137361175135,
+ "mean_feedback_forward_relative_error": 0.2350148629201086,
+ "min_feedback_forward_cosine": 0.9389908909797668
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 13294,
+ "train_examples": 50000,
+ "train_loss": 0.2878492239189148
+ },
+ {
+ "epoch": 35,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.3398529887199402,
+ "mean_feedback_forward_cosine": 0.974117658640209,
+ "mean_feedback_forward_relative_error": 0.22552741554222608,
+ "min_feedback_forward_cosine": 0.9440118074417114
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 13685,
+ "train_examples": 50000,
+ "train_loss": 0.2892651585960388
+ },
+ {
+ "epoch": 36,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.32580846548080444,
+ "mean_feedback_forward_cosine": 0.9760581129475644,
+ "mean_feedback_forward_relative_error": 0.2168219320868191,
+ "min_feedback_forward_cosine": 0.9484668970108032
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 14076,
+ "train_examples": 50000,
+ "train_loss": 0.28420805393218995
+ },
+ {
+ "epoch": 37,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.31265878677368164,
+ "mean_feedback_forward_cosine": 0.9778313511296323,
+ "mean_feedback_forward_relative_error": 0.20859109688746302,
+ "min_feedback_forward_cosine": 0.952483057975769
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 14467,
+ "train_examples": 50000,
+ "train_loss": 0.27960984511375425
+ },
+ {
+ "epoch": 38,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.2996475398540497,
+ "mean_feedback_forward_cosine": 0.9795032802381014,
+ "mean_feedback_forward_relative_error": 0.20050902821515737,
+ "min_feedback_forward_cosine": 0.9563074111938477
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 14858,
+ "train_examples": 50000,
+ "train_loss": 0.27376583651542663
+ },
+ {
+ "epoch": 39,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.28715112805366516,
+ "mean_feedback_forward_cosine": 0.9810729873807806,
+ "mean_feedback_forward_relative_error": 0.19261449967560015,
+ "min_feedback_forward_cosine": 0.9598467350006104
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 15249,
+ "train_examples": 50000,
+ "train_loss": 0.2761541701221466
+ },
+ {
+ "epoch": 40,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.27562326192855835,
+ "mean_feedback_forward_cosine": 0.9825075017778497,
+ "mean_feedback_forward_relative_error": 0.1851246988302783,
+ "min_feedback_forward_cosine": 0.9629642963409424
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 15640,
+ "train_examples": 50000,
+ "train_loss": 0.27317368170261386
+ },
+ {
+ "epoch": 41,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.26422426104545593,
+ "mean_feedback_forward_cosine": 0.9838146661457262,
+ "mean_feedback_forward_relative_error": 0.17799867846463857,
+ "min_feedback_forward_cosine": 0.9659215211868286
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 16031,
+ "train_examples": 50000,
+ "train_loss": 0.27047356218338015
+ },
+ {
+ "epoch": 42,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.25380969047546387,
+ "mean_feedback_forward_cosine": 0.9850201951830011,
+ "mean_feedback_forward_relative_error": 0.17117400310541453,
+ "min_feedback_forward_cosine": 0.9685237407684326
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 16422,
+ "train_examples": 50000,
+ "train_loss": 0.26298034185409547
+ },
+ {
+ "epoch": 43,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.2434985190629959,
+ "mean_feedback_forward_cosine": 0.9861482789641932,
+ "mean_feedback_forward_relative_error": 0.16458421906358317,
+ "min_feedback_forward_cosine": 0.9710122346878052
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 16813,
+ "train_examples": 50000,
+ "train_loss": 0.2653024362277985
+ },
+ {
+ "epoch": 44,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.23392316699028015,
+ "mean_feedback_forward_cosine": 0.9871551331720854,
+ "mean_feedback_forward_relative_error": 0.1584414451530105,
+ "min_feedback_forward_cosine": 0.9732286930084229
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 17204,
+ "train_examples": 50000,
+ "train_loss": 0.259649914932251
+ },
+ {
+ "epoch": 45,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.22451873123645782,
+ "mean_feedback_forward_cosine": 0.9881107054258648,
+ "mean_feedback_forward_relative_error": 0.15240878258880816,
+ "min_feedback_forward_cosine": 0.975327730178833
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 17595,
+ "train_examples": 50000,
+ "train_loss": 0.2577659420776367
+ },
+ {
+ "epoch": 46,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.21546627581119537,
+ "mean_feedback_forward_cosine": 0.9889990436403375,
+ "mean_feedback_forward_relative_error": 0.1465405433585769,
+ "min_feedback_forward_cosine": 0.9772522449493408
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 17986,
+ "train_examples": 50000,
+ "train_loss": 0.25411363348960875
+ },
+ {
+ "epoch": 47,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.20682618021965027,
+ "mean_feedback_forward_cosine": 0.9898229774675871,
+ "mean_feedback_forward_relative_error": 0.14092829999955078,
+ "min_feedback_forward_cosine": 0.9790107011795044
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 18377,
+ "train_examples": 50000,
+ "train_loss": 0.25334115114212036
+ },
+ {
+ "epoch": 48,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.19855043292045593,
+ "mean_feedback_forward_cosine": 0.9905795423608077,
+ "mean_feedback_forward_relative_error": 0.1355620422645619,
+ "min_feedback_forward_cosine": 0.9806437492370605
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 18768,
+ "train_examples": 50000,
+ "train_loss": 0.25010806784152984
+ },
+ {
+ "epoch": 49,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.19042891263961792,
+ "mean_feedback_forward_cosine": 0.9912930507408945,
+ "mean_feedback_forward_relative_error": 0.13029807453092776,
+ "min_feedback_forward_cosine": 0.9821843504905701
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 19159,
+ "train_examples": 50000,
+ "train_loss": 0.2482856145954132
+ },
+ {
+ "epoch": 50,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.18241143226623535,
+ "mean_feedback_forward_cosine": 0.9919578276182476,
+ "mean_feedback_forward_relative_error": 0.1252188225717921,
+ "min_feedback_forward_cosine": 0.9836432933807373
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 19550,
+ "train_examples": 50000,
+ "train_loss": 0.25066015458106994
+ },
+ {
+ "epoch": 51,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.17519184947013855,
+ "mean_feedback_forward_cosine": 0.9925572589824074,
+ "mean_feedback_forward_relative_error": 0.12043296800632226,
+ "min_feedback_forward_cosine": 0.9848968386650085
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 19941,
+ "train_examples": 50000,
+ "train_loss": 0.24143317845344545
+ },
+ {
+ "epoch": 52,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.1679893434047699,
+ "mean_feedback_forward_cosine": 0.9931172289346394,
+ "mean_feedback_forward_relative_error": 0.11580108321811024,
+ "min_feedback_forward_cosine": 0.9861008524894714
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 20332,
+ "train_examples": 50000,
+ "train_loss": 0.24608960815429687
+ },
+ {
+ "epoch": 53,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.16124595701694489,
+ "mean_feedback_forward_cosine": 0.993638559391624,
+ "mean_feedback_forward_relative_error": 0.11130892171671516,
+ "min_feedback_forward_cosine": 0.9871878027915955
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 20723,
+ "train_examples": 50000,
+ "train_loss": 0.24206441249847413
+ },
+ {
+ "epoch": 54,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.15466845035552979,
+ "mean_feedback_forward_cosine": 0.9941155659525018,
+ "mean_feedback_forward_relative_error": 0.10703967903789721,
+ "min_feedback_forward_cosine": 0.9882041811943054
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 21114,
+ "train_examples": 50000,
+ "train_loss": 0.2399489393520355
+ },
+ {
+ "epoch": 55,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.14854159951210022,
+ "mean_feedback_forward_cosine": 0.9945565273887232,
+ "mean_feedback_forward_relative_error": 0.10293684252782871,
+ "min_feedback_forward_cosine": 0.9891126155853271
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 21505,
+ "train_examples": 50000,
+ "train_loss": 0.23437146086215974
+ },
+ {
+ "epoch": 56,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.14255544543266296,
+ "mean_feedback_forward_cosine": 0.9949761534992018,
+ "mean_feedback_forward_relative_error": 0.09887950259603952,
+ "min_feedback_forward_cosine": 0.9899699687957764
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 21896,
+ "train_examples": 50000,
+ "train_loss": 0.23884480169296265
+ },
+ {
+ "epoch": 57,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.13674475252628326,
+ "mean_feedback_forward_cosine": 0.9953545771147075,
+ "mean_feedback_forward_relative_error": 0.09507158614302937,
+ "min_feedback_forward_cosine": 0.990766167640686
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 22287,
+ "train_examples": 50000,
+ "train_loss": 0.2372421080970764
+ },
+ {
+ "epoch": 58,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.13124753534793854,
+ "mean_feedback_forward_cosine": 0.9957059057135331,
+ "mean_feedback_forward_relative_error": 0.09138967547761767,
+ "min_feedback_forward_cosine": 0.9914892315864563
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 22678,
+ "train_examples": 50000,
+ "train_loss": 0.23350110571861268
+ },
+ {
+ "epoch": 59,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.12596161663532257,
+ "mean_feedback_forward_cosine": 0.9960281158748426,
+ "mean_feedback_forward_relative_error": 0.08788347479544188,
+ "min_feedback_forward_cosine": 0.9921553730964661
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 23069,
+ "train_examples": 50000,
+ "train_loss": 0.23144394822120667
+ },
+ {
+ "epoch": 60,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.12090470641851425,
+ "mean_feedback_forward_cosine": 0.9963286173971075,
+ "mean_feedback_forward_relative_error": 0.08449560776352882,
+ "min_feedback_forward_cosine": 0.9927682876586914
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 23460,
+ "train_examples": 50000,
+ "train_loss": 0.2364779268360138
+ },
+ {
+ "epoch": 61,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.11604072153568268,
+ "mean_feedback_forward_cosine": 0.996597525320555,
+ "mean_feedback_forward_relative_error": 0.08133108815864513,
+ "min_feedback_forward_cosine": 0.993336021900177
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 23851,
+ "train_examples": 50000,
+ "train_loss": 0.22808763239860536
+ },
+ {
+ "epoch": 62,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.11152151972055435,
+ "mean_feedback_forward_cosine": 0.9968531445453042,
+ "mean_feedback_forward_relative_error": 0.07820960898932658,
+ "min_feedback_forward_cosine": 0.9938417673110962
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 24242,
+ "train_examples": 50000,
+ "train_loss": 0.2266447284221649
+ },
+ {
+ "epoch": 63,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.10693155974149704,
+ "mean_feedback_forward_cosine": 0.9970944115990087,
+ "mean_feedback_forward_relative_error": 0.07515961716049596,
+ "min_feedback_forward_cosine": 0.994335412979126
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 24633,
+ "train_examples": 50000,
+ "train_loss": 0.23168384791374205
+ },
+ {
+ "epoch": 64,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.10286647826433182,
+ "mean_feedback_forward_cosine": 0.9973037556598061,
+ "mean_feedback_forward_relative_error": 0.0723838446367728,
+ "min_feedback_forward_cosine": 0.9947562217712402
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 25024,
+ "train_examples": 50000,
+ "train_loss": 0.21967432646751403
+ },
+ {
+ "epoch": 65,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.09874574095010757,
+ "mean_feedback_forward_cosine": 0.9975028916409141,
+ "mean_feedback_forward_relative_error": 0.06964617242154322,
+ "min_feedback_forward_cosine": 0.9951664209365845
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 25415,
+ "train_examples": 50000,
+ "train_loss": 0.22276782621383667
+ },
+ {
+ "epoch": 66,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.09475722908973694,
+ "mean_feedback_forward_cosine": 0.9976867186395746,
+ "mean_feedback_forward_relative_error": 0.06702639535069466,
+ "min_feedback_forward_cosine": 0.9955471754074097
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 25806,
+ "train_examples": 50000,
+ "train_loss": 0.2208927348947525
+ },
+ {
+ "epoch": 67,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.09098617732524872,
+ "mean_feedback_forward_cosine": 0.9978609085083008,
+ "mean_feedback_forward_relative_error": 0.06444480201523554,
+ "min_feedback_forward_cosine": 0.9958927631378174
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 26197,
+ "train_examples": 50000,
+ "train_loss": 0.21977276565551757
+ },
+ {
+ "epoch": 68,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.08740058541297913,
+ "mean_feedback_forward_cosine": 0.9980220292743883,
+ "mean_feedback_forward_relative_error": 0.06196602424116511,
+ "min_feedback_forward_cosine": 0.9962090253829956
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 26588,
+ "train_examples": 50000,
+ "train_loss": 0.21752099536895753
+ },
+ {
+ "epoch": 69,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.08383046090602875,
+ "mean_feedback_forward_cosine": 0.9981727349130731,
+ "mean_feedback_forward_relative_error": 0.05956339208703292,
+ "min_feedback_forward_cosine": 0.9965111017227173
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 26979,
+ "train_examples": 50000,
+ "train_loss": 0.22395552233695984
+ },
+ {
+ "epoch": 70,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0804688110947609,
+ "mean_feedback_forward_cosine": 0.9983122599752325,
+ "mean_feedback_forward_relative_error": 0.05724060035457736,
+ "min_feedback_forward_cosine": 0.996785044670105
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 27370,
+ "train_examples": 50000,
+ "train_loss": 0.22269117835998536
+ },
+ {
+ "epoch": 71,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.07737977057695389,
+ "mean_feedback_forward_cosine": 0.998433928740652,
+ "mean_feedback_forward_relative_error": 0.05512593872845173,
+ "min_feedback_forward_cosine": 0.9970258474349976
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 27761,
+ "train_examples": 50000,
+ "train_loss": 0.21499397795677186
+ },
+ {
+ "epoch": 72,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.07427549362182617,
+ "mean_feedback_forward_cosine": 0.9985538501488535,
+ "mean_feedback_forward_relative_error": 0.05297956811754327,
+ "min_feedback_forward_cosine": 0.9972590208053589
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 28152,
+ "train_examples": 50000,
+ "train_loss": 0.21951112261772157
+ },
+ {
+ "epoch": 73,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.07133505493402481,
+ "mean_feedback_forward_cosine": 0.9986621141433716,
+ "mean_feedback_forward_relative_error": 0.0509505889525539,
+ "min_feedback_forward_cosine": 0.9974708557128906
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 28543,
+ "train_examples": 50000,
+ "train_loss": 0.2146762759590149
+ },
+ {
+ "epoch": 74,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.068573497235775,
+ "mean_feedback_forward_cosine": 0.9987624381717882,
+ "mean_feedback_forward_relative_error": 0.04900112885393595,
+ "min_feedback_forward_cosine": 0.9976617693901062
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 28934,
+ "train_examples": 50000,
+ "train_loss": 0.21213805459976195
+ },
+ {
+ "epoch": 75,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.06581214815378189,
+ "mean_feedback_forward_cosine": 0.9988550763381155,
+ "mean_feedback_forward_relative_error": 0.04713135840077149,
+ "min_feedback_forward_cosine": 0.9978461861610413
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 29325,
+ "train_examples": 50000,
+ "train_loss": 0.21292313886642456
+ },
+ {
+ "epoch": 76,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.06323916465044022,
+ "mean_feedback_forward_cosine": 0.9989409383974577,
+ "mean_feedback_forward_relative_error": 0.04532517061421746,
+ "min_feedback_forward_cosine": 0.998010516166687
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 29716,
+ "train_examples": 50000,
+ "train_loss": 0.21277331172943115
+ },
+ {
+ "epoch": 77,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.06074615940451622,
+ "mean_feedback_forward_cosine": 0.9990207207830328,
+ "mean_feedback_forward_relative_error": 0.04358574346100029,
+ "min_feedback_forward_cosine": 0.9981639385223389
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 30107,
+ "train_examples": 50000,
+ "train_loss": 0.20898236991882324
+ },
+ {
+ "epoch": 78,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.05832308903336525,
+ "mean_feedback_forward_cosine": 0.9990944454544469,
+ "mean_feedback_forward_relative_error": 0.0419123242364118,
+ "min_feedback_forward_cosine": 0.9983067512512207
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 30498,
+ "train_examples": 50000,
+ "train_loss": 0.21116136218070983
+ },
+ {
+ "epoch": 79,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0559818260371685,
+ "mean_feedback_forward_cosine": 0.9991625045475206,
+ "mean_feedback_forward_relative_error": 0.04030360301074229,
+ "min_feedback_forward_cosine": 0.9984397292137146
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 30889,
+ "train_examples": 50000,
+ "train_loss": 0.21172104592323304
+ },
+ {
+ "epoch": 80,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.05379173532128334,
+ "mean_feedback_forward_cosine": 0.9992254564636632,
+ "mean_feedback_forward_relative_error": 0.03875644593254516,
+ "min_feedback_forward_cosine": 0.9985591173171997
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 31280,
+ "train_examples": 50000,
+ "train_loss": 0.20921241371154786
+ },
+ {
+ "epoch": 81,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.051668789237737656,
+ "mean_feedback_forward_cosine": 0.9992847285772625,
+ "mean_feedback_forward_relative_error": 0.037242337767230835,
+ "min_feedback_forward_cosine": 0.9986700415611267
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 31671,
+ "train_examples": 50000,
+ "train_loss": 0.20510462395191192
+ },
+ {
+ "epoch": 82,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.04965691268444061,
+ "mean_feedback_forward_cosine": 0.9993384574588976,
+ "mean_feedback_forward_relative_error": 0.03581918712313238,
+ "min_feedback_forward_cosine": 0.9987714886665344
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 32062,
+ "train_examples": 50000,
+ "train_loss": 0.2070067384815216
+ },
+ {
+ "epoch": 83,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.047659263014793396,
+ "mean_feedback_forward_cosine": 0.9993902633064672,
+ "mean_feedback_forward_relative_error": 0.034390744164978206,
+ "min_feedback_forward_cosine": 0.9988679885864258
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 32453,
+ "train_examples": 50000,
+ "train_loss": 0.20999451464653016
+ },
+ {
+ "epoch": 84,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.045759331434965134,
+ "mean_feedback_forward_cosine": 0.9994356694974398,
+ "mean_feedback_forward_relative_error": 0.03308229128781118,
+ "min_feedback_forward_cosine": 0.9989561438560486
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 32844,
+ "train_examples": 50000,
+ "train_loss": 0.20893828397274017
+ },
+ {
+ "epoch": 85,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.04397435113787651,
+ "mean_feedback_forward_cosine": 0.9994768374844601,
+ "mean_feedback_forward_relative_error": 0.03184907133445928,
+ "min_feedback_forward_cosine": 0.9990359544754028
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 33235,
+ "train_examples": 50000,
+ "train_loss": 0.20038848754405975
+ },
+ {
+ "epoch": 86,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.04222673177719116,
+ "mean_feedback_forward_cosine": 0.999517095716376,
+ "mean_feedback_forward_relative_error": 0.03060037922114134,
+ "min_feedback_forward_cosine": 0.9991111159324646
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 33626,
+ "train_examples": 50000,
+ "train_loss": 0.2064295777606964
+ },
+ {
+ "epoch": 87,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.04057107865810394,
+ "mean_feedback_forward_cosine": 0.9995530153575697,
+ "mean_feedback_forward_relative_error": 0.029441306073414653,
+ "min_feedback_forward_cosine": 0.9991794228553772
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 34017,
+ "train_examples": 50000,
+ "train_loss": 0.20408623010635377
+ },
+ {
+ "epoch": 88,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03902265802025795,
+ "mean_feedback_forward_cosine": 0.9995857100737723,
+ "mean_feedback_forward_relative_error": 0.02834192138949507,
+ "min_feedback_forward_cosine": 0.9992407560348511
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 34408,
+ "train_examples": 50000,
+ "train_loss": 0.20103353147029876
+ },
+ {
+ "epoch": 89,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.037542738020420074,
+ "mean_feedback_forward_cosine": 0.9996166323360643,
+ "mean_feedback_forward_relative_error": 0.027261023654749517,
+ "min_feedback_forward_cosine": 0.9992972612380981
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 34799,
+ "train_examples": 50000,
+ "train_loss": 0.19745080633163453
+ },
+ {
+ "epoch": 90,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0360117070376873,
+ "mean_feedback_forward_cosine": 0.9996461774173536,
+ "mean_feedback_forward_relative_error": 0.02619106748974637,
+ "min_feedback_forward_cosine": 0.9993532299995422
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 35190,
+ "train_examples": 50000,
+ "train_loss": 0.20549977100372313
+ },
+ {
+ "epoch": 91,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03458224982023239,
+ "mean_feedback_forward_cosine": 0.9996738026016637,
+ "mean_feedback_forward_relative_error": 0.025146570889965483,
+ "min_feedback_forward_cosine": 0.9994035959243774
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 35581,
+ "train_examples": 50000,
+ "train_loss": 0.1986873722934723
+ },
+ {
+ "epoch": 92,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03324142470955849,
+ "mean_feedback_forward_cosine": 0.9996980867887798,
+ "mean_feedback_forward_relative_error": 0.024191624141837422,
+ "min_feedback_forward_cosine": 0.9994487762451172
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 35972,
+ "train_examples": 50000,
+ "train_loss": 0.19804216557025908
+ },
+ {
+ "epoch": 93,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03191523626446724,
+ "mean_feedback_forward_cosine": 0.9997214737691378,
+ "mean_feedback_forward_relative_error": 0.023236432455872234,
+ "min_feedback_forward_cosine": 0.999491810798645
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 36363,
+ "train_examples": 50000,
+ "train_loss": 0.20227410102844237
+ },
+ {
+ "epoch": 94,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.030632607638835907,
+ "mean_feedback_forward_cosine": 0.9997428404657465,
+ "mean_feedback_forward_relative_error": 0.02232741314525667,
+ "min_feedback_forward_cosine": 0.9995319247245789
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 36754,
+ "train_examples": 50000,
+ "train_loss": 0.20241206522464752
+ },
+ {
+ "epoch": 95,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.029394421726465225,
+ "mean_feedback_forward_cosine": 0.9997620331613641,
+ "mean_feedback_forward_relative_error": 0.02147653875382323,
+ "min_feedback_forward_cosine": 0.9995689392089844
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 37145,
+ "train_examples": 50000,
+ "train_loss": 0.20295502564907075
+ },
+ {
+ "epoch": 96,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.028264634311199188,
+ "mean_feedback_forward_cosine": 0.9997799741594415,
+ "mean_feedback_forward_relative_error": 0.020652579584796178,
+ "min_feedback_forward_cosine": 0.9996013045310974
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 37536,
+ "train_examples": 50000,
+ "train_loss": 0.19767562034606934
+ },
+ {
+ "epoch": 97,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02715025283396244,
+ "mean_feedback_forward_cosine": 0.9997962775983309,
+ "mean_feedback_forward_relative_error": 0.019869652380676645,
+ "min_feedback_forward_cosine": 0.9996321201324463
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 37927,
+ "train_examples": 50000,
+ "train_loss": 0.19347093383550643
+ },
+ {
+ "epoch": 98,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02612755261361599,
+ "mean_feedback_forward_cosine": 0.9998112007191307,
+ "mean_feedback_forward_relative_error": 0.019126742195926215,
+ "min_feedback_forward_cosine": 0.9996593594551086
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 38318,
+ "train_examples": 50000,
+ "train_loss": 0.19226393731117247
+ },
+ {
+ "epoch": 99,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.025075765326619148,
+ "mean_feedback_forward_cosine": 0.9998256250431663,
+ "mean_feedback_forward_relative_error": 0.01837975173992546,
+ "min_feedback_forward_cosine": 0.9996861815452576
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 38709,
+ "train_examples": 50000,
+ "train_loss": 0.19686378298282622
+ },
+ {
+ "epoch": 100,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.024072103202342987,
+ "mean_feedback_forward_cosine": 0.9998391396120975,
+ "mean_feedback_forward_relative_error": 0.017656986258531873,
+ "min_feedback_forward_cosine": 0.9997107982635498
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 39100,
+ "train_examples": 50000,
+ "train_loss": 0.1982896047115326
+ },
+ {
+ "epoch": 101,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02406390942633152,
+ "mean_feedback_forward_cosine": 0.9998392996035124,
+ "mean_feedback_forward_relative_error": 0.01764692640618274,
+ "min_feedback_forward_cosine": 0.9997109174728394
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 39491,
+ "train_examples": 50000,
+ "train_loss": 0.12299238201379777
+ },
+ {
+ "epoch": 102,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.024059180170297623,
+ "mean_feedback_forward_cosine": 0.9998393686194169,
+ "mean_feedback_forward_relative_error": 0.017639402226593933,
+ "min_feedback_forward_cosine": 0.9997110366821289
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 39882,
+ "train_examples": 50000,
+ "train_loss": 0.09884670098781585
+ },
+ {
+ "epoch": 103,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02405465766787529,
+ "mean_feedback_forward_cosine": 0.9998394313611483,
+ "mean_feedback_forward_relative_error": 0.017632367064882266,
+ "min_feedback_forward_cosine": 0.999711275100708
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 40273,
+ "train_examples": 50000,
+ "train_loss": 0.08819357568740845
+ },
+ {
+ "epoch": 104,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.024050189182162285,
+ "mean_feedback_forward_cosine": 0.9998395191995721,
+ "mean_feedback_forward_relative_error": 0.017625431652720038,
+ "min_feedback_forward_cosine": 0.999711275100708
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 40664,
+ "train_examples": 50000,
+ "train_loss": 0.08101649174511433
+ },
+ {
+ "epoch": 105,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02404564991593361,
+ "mean_feedback_forward_cosine": 0.9998396415459482,
+ "mean_feedback_forward_relative_error": 0.017618758236303142,
+ "min_feedback_forward_cosine": 0.9997114539146423
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 41055,
+ "train_examples": 50000,
+ "train_loss": 0.0770462325143814
+ },
+ {
+ "epoch": 106,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.024041302502155304,
+ "mean_feedback_forward_cosine": 0.9998396980135065,
+ "mean_feedback_forward_relative_error": 0.017612302185673463,
+ "min_feedback_forward_cosine": 0.9997115135192871
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 41446,
+ "train_examples": 50000,
+ "train_loss": 0.07072865057945252
+ },
+ {
+ "epoch": 107,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.024036701768636703,
+ "mean_feedback_forward_cosine": 0.99983979526319,
+ "mean_feedback_forward_relative_error": 0.01760593789482587,
+ "min_feedback_forward_cosine": 0.9997116327285767
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 41837,
+ "train_examples": 50000,
+ "train_loss": 0.06999157873630524
+ },
+ {
+ "epoch": 108,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.024032164365053177,
+ "mean_feedback_forward_cosine": 0.9998398423194885,
+ "mean_feedback_forward_relative_error": 0.017599599656501885,
+ "min_feedback_forward_cosine": 0.9997117519378662
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 42228,
+ "train_examples": 50000,
+ "train_loss": 0.06534900736093521
+ },
+ {
+ "epoch": 109,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02402760460972786,
+ "mean_feedback_forward_cosine": 0.9998399458433452,
+ "mean_feedback_forward_relative_error": 0.017593359721726494,
+ "min_feedback_forward_cosine": 0.9997119307518005
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 42619,
+ "train_examples": 50000,
+ "train_loss": 0.06322124888181686
+ },
+ {
+ "epoch": 110,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.024022996425628662,
+ "mean_feedback_forward_cosine": 0.9998400179963363,
+ "mean_feedback_forward_relative_error": 0.017587167137351475,
+ "min_feedback_forward_cosine": 0.9997119903564453
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 43010,
+ "train_examples": 50000,
+ "train_loss": 0.06139645607709884
+ },
+ {
+ "epoch": 111,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.024018622934818268,
+ "mean_feedback_forward_cosine": 0.9998400901493273,
+ "mean_feedback_forward_relative_error": 0.01758105947489017,
+ "min_feedback_forward_cosine": 0.9997121095657349
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 43401,
+ "train_examples": 50000,
+ "train_loss": 0.05795342464208603
+ },
+ {
+ "epoch": 112,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.024014070630073547,
+ "mean_feedback_forward_cosine": 0.9998401873990109,
+ "mean_feedback_forward_relative_error": 0.017574954949515432,
+ "min_feedback_forward_cosine": 0.9997121691703796
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 43792,
+ "train_examples": 50000,
+ "train_loss": 0.0559422425699234
+ },
+ {
+ "epoch": 113,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.024009378626942635,
+ "mean_feedback_forward_cosine": 0.9998402689632616,
+ "mean_feedback_forward_relative_error": 0.017568798809263268,
+ "min_feedback_forward_cosine": 0.999712347984314
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 44183,
+ "train_examples": 50000,
+ "train_loss": 0.055113451805114747
+ },
+ {
+ "epoch": 114,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02400479093194008,
+ "mean_feedback_forward_cosine": 0.9998403473904258,
+ "mean_feedback_forward_relative_error": 0.017562687029375842,
+ "min_feedback_forward_cosine": 0.9997124075889587
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 44574,
+ "train_examples": 50000,
+ "train_loss": 0.052615987684726716
+ },
+ {
+ "epoch": 115,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.024000002071261406,
+ "mean_feedback_forward_cosine": 0.9998404195434168,
+ "mean_feedback_forward_relative_error": 0.017556504248396346,
+ "min_feedback_forward_cosine": 0.9997124671936035
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 44965,
+ "train_examples": 50000,
+ "train_loss": 0.051486412073373795
+ },
+ {
+ "epoch": 116,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023995356634259224,
+ "mean_feedback_forward_cosine": 0.9998405293414467,
+ "mean_feedback_forward_relative_error": 0.017550376636025152,
+ "min_feedback_forward_cosine": 0.9997127056121826
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 45356,
+ "train_examples": 50000,
+ "train_loss": 0.04837573773741722
+ },
+ {
+ "epoch": 117,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02399068884551525,
+ "mean_feedback_forward_cosine": 0.9998405920831781,
+ "mean_feedback_forward_relative_error": 0.01754426519925657,
+ "min_feedback_forward_cosine": 0.9997127652168274
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 45747,
+ "train_examples": 50000,
+ "train_loss": 0.047202864929437635
+ },
+ {
+ "epoch": 118,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023985648527741432,
+ "mean_feedback_forward_cosine": 0.9998407018812079,
+ "mean_feedback_forward_relative_error": 0.0175379439943323,
+ "min_feedback_forward_cosine": 0.9997129440307617
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 46138,
+ "train_examples": 50000,
+ "train_loss": 0.04749902138710022
+ },
+ {
+ "epoch": 119,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023980626836419106,
+ "mean_feedback_forward_cosine": 0.9998407897196318,
+ "mean_feedback_forward_relative_error": 0.017531843120722396,
+ "min_feedback_forward_cosine": 0.9997129440307617
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 46529,
+ "train_examples": 50000,
+ "train_loss": 0.04516030862092972
+ },
+ {
+ "epoch": 120,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023976139724254608,
+ "mean_feedback_forward_cosine": 0.99984084618719,
+ "mean_feedback_forward_relative_error": 0.01752598497918562,
+ "min_feedback_forward_cosine": 0.999713122844696
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 46920,
+ "train_examples": 50000,
+ "train_loss": 0.04241939604222775
+ },
+ {
+ "epoch": 121,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02397124096751213,
+ "mean_feedback_forward_cosine": 0.9998409371627005,
+ "mean_feedback_forward_relative_error": 0.01751989868812655,
+ "min_feedback_forward_cosine": 0.9997131824493408
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 47311,
+ "train_examples": 50000,
+ "train_loss": 0.042379607338905334
+ },
+ {
+ "epoch": 122,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02396647445857525,
+ "mean_feedback_forward_cosine": 0.9998410093156915,
+ "mean_feedback_forward_relative_error": 0.017513908837970934,
+ "min_feedback_forward_cosine": 0.9997134208679199
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 47702,
+ "train_examples": 50000,
+ "train_loss": 0.041688693816661834
+ },
+ {
+ "epoch": 123,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023961400613188744,
+ "mean_feedback_forward_cosine": 0.9998411034282885,
+ "mean_feedback_forward_relative_error": 0.017507812302363545,
+ "min_feedback_forward_cosine": 0.9997134208679199
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 48093,
+ "train_examples": 50000,
+ "train_loss": 0.04086528647184372
+ },
+ {
+ "epoch": 124,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023956535384058952,
+ "mean_feedback_forward_cosine": 0.9998412038150587,
+ "mean_feedback_forward_relative_error": 0.017501762038783023,
+ "min_feedback_forward_cosine": 0.9997134804725647
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 48484,
+ "train_examples": 50000,
+ "train_loss": 0.03863505096554756
+ },
+ {
+ "epoch": 125,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02395145781338215,
+ "mean_feedback_forward_cosine": 0.9998413073389154,
+ "mean_feedback_forward_relative_error": 0.017495613153043547,
+ "min_feedback_forward_cosine": 0.9997137784957886
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 48875,
+ "train_examples": 50000,
+ "train_loss": 0.039321243322193625
+ },
+ {
+ "epoch": 126,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02394598349928856,
+ "mean_feedback_forward_cosine": 0.9998413889031661,
+ "mean_feedback_forward_relative_error": 0.01748918663514288,
+ "min_feedback_forward_cosine": 0.9997138977050781
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 49266,
+ "train_examples": 50000,
+ "train_loss": 0.03911113297462463
+ },
+ {
+ "epoch": 127,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02394097112119198,
+ "mean_feedback_forward_cosine": 0.9998414798786766,
+ "mean_feedback_forward_relative_error": 0.01748289189938652,
+ "min_feedback_forward_cosine": 0.9997137784957886
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 49657,
+ "train_examples": 50000,
+ "train_loss": 0.036497448803186415
+ },
+ {
+ "epoch": 128,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023935630917549133,
+ "mean_feedback_forward_cosine": 0.9998415802654467,
+ "mean_feedback_forward_relative_error": 0.017476465430502828,
+ "min_feedback_forward_cosine": 0.9997141361236572
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 50048,
+ "train_examples": 50000,
+ "train_loss": 0.03716952535510063
+ },
+ {
+ "epoch": 129,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02393004484474659,
+ "mean_feedback_forward_cosine": 0.9998416649667841,
+ "mean_feedback_forward_relative_error": 0.01747012307475272,
+ "min_feedback_forward_cosine": 0.9997142553329468
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 50439,
+ "train_examples": 50000,
+ "train_loss": 0.035937383514642714
+ },
+ {
+ "epoch": 130,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023924510926008224,
+ "mean_feedback_forward_cosine": 0.9998417653535542,
+ "mean_feedback_forward_relative_error": 0.01746368145962295,
+ "min_feedback_forward_cosine": 0.9997143149375916
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 50830,
+ "train_examples": 50000,
+ "train_loss": 0.03566022593140602
+ },
+ {
+ "epoch": 131,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023919150233268738,
+ "mean_feedback_forward_cosine": 0.9998418406436318,
+ "mean_feedback_forward_relative_error": 0.017457206978609686,
+ "min_feedback_forward_cosine": 0.9997143745422363
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 51221,
+ "train_examples": 50000,
+ "train_loss": 0.034828656885623935
+ },
+ {
+ "epoch": 132,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023913824930787086,
+ "mean_feedback_forward_cosine": 0.9998419661270944,
+ "mean_feedback_forward_relative_error": 0.0174507660497176,
+ "min_feedback_forward_cosine": 0.9997146725654602
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 51612,
+ "train_examples": 50000,
+ "train_loss": 0.033103851650431754
+ },
+ {
+ "epoch": 133,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02390843629837036,
+ "mean_feedback_forward_cosine": 0.9998420539655184,
+ "mean_feedback_forward_relative_error": 0.017444494842110497,
+ "min_feedback_forward_cosine": 0.9997147917747498
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 52003,
+ "train_examples": 50000,
+ "train_loss": 0.032778032338619235
+ },
+ {
+ "epoch": 134,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023902934044599533,
+ "mean_feedback_forward_cosine": 0.9998421574893751,
+ "mean_feedback_forward_relative_error": 0.017437889265190614,
+ "min_feedback_forward_cosine": 0.9997148513793945
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 52394,
+ "train_examples": 50000,
+ "train_loss": 0.03193912905871868
+ },
+ {
+ "epoch": 135,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023897364735603333,
+ "mean_feedback_forward_cosine": 0.9998422547390586,
+ "mean_feedback_forward_relative_error": 0.017431417480111122,
+ "min_feedback_forward_cosine": 0.9997149705886841
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 52785,
+ "train_examples": 50000,
+ "train_loss": 0.0319919503903389
+ },
+ {
+ "epoch": 136,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02389141172170639,
+ "mean_feedback_forward_cosine": 0.9998423739483482,
+ "mean_feedback_forward_relative_error": 0.01742457061711895,
+ "min_feedback_forward_cosine": 0.9997152090072632
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 53176,
+ "train_examples": 50000,
+ "train_loss": 0.03163765184640884
+ },
+ {
+ "epoch": 137,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023885833099484444,
+ "mean_feedback_forward_cosine": 0.9998424680609452,
+ "mean_feedback_forward_relative_error": 0.017417882520117257,
+ "min_feedback_forward_cosine": 0.9997152090072632
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 53567,
+ "train_examples": 50000,
+ "train_loss": 0.030665701761841774
+ },
+ {
+ "epoch": 138,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02388034388422966,
+ "mean_feedback_forward_cosine": 0.9998425684477154,
+ "mean_feedback_forward_relative_error": 0.017411238587412396,
+ "min_feedback_forward_cosine": 0.999715268611908
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 53958,
+ "train_examples": 50000,
+ "train_loss": 0.029491460859775543
+ },
+ {
+ "epoch": 139,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023874782025814056,
+ "mean_feedback_forward_cosine": 0.9998426719715721,
+ "mean_feedback_forward_relative_error": 0.017404530834602683,
+ "min_feedback_forward_cosine": 0.9997155666351318
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 54349,
+ "train_examples": 50000,
+ "train_loss": 0.0298191662889719
+ },
+ {
+ "epoch": 140,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02386835590004921,
+ "mean_feedback_forward_cosine": 0.9998427692212557,
+ "mean_feedback_forward_relative_error": 0.0173971701511427,
+ "min_feedback_forward_cosine": 0.9997156858444214
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 54740,
+ "train_examples": 50000,
+ "train_loss": 0.03218640671610832
+ },
+ {
+ "epoch": 141,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02386237494647503,
+ "mean_feedback_forward_cosine": 0.9998428821563721,
+ "mean_feedback_forward_relative_error": 0.017390342747890634,
+ "min_feedback_forward_cosine": 0.9997157454490662
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 55131,
+ "train_examples": 50000,
+ "train_loss": 0.029018398514389993
+ },
+ {
+ "epoch": 142,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02385648339986801,
+ "mean_feedback_forward_cosine": 0.9998429794060556,
+ "mean_feedback_forward_relative_error": 0.01738346966081544,
+ "min_feedback_forward_cosine": 0.9997159838676453
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 55522,
+ "train_examples": 50000,
+ "train_loss": 0.029316624326705933
+ },
+ {
+ "epoch": 143,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023850666359066963,
+ "mean_feedback_forward_cosine": 0.9998430766557392,
+ "mean_feedback_forward_relative_error": 0.01737667786839761,
+ "min_feedback_forward_cosine": 0.99971604347229
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 55913,
+ "train_examples": 50000,
+ "train_loss": 0.02739346834808588
+ },
+ {
+ "epoch": 144,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023844413459300995,
+ "mean_feedback_forward_cosine": 0.9998432429213273,
+ "mean_feedback_forward_relative_error": 0.01736944132043343,
+ "min_feedback_forward_cosine": 0.9997162818908691
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 56304,
+ "train_examples": 50000,
+ "train_loss": 0.028192082404494286
+ },
+ {
+ "epoch": 145,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023838169872760773,
+ "mean_feedback_forward_cosine": 0.9998433307597512,
+ "mean_feedback_forward_relative_error": 0.01736224898578305,
+ "min_feedback_forward_cosine": 0.9997164011001587
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 56695,
+ "train_examples": 50000,
+ "train_loss": 0.0277932139891386
+ },
+ {
+ "epoch": 146,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023832306265830994,
+ "mean_feedback_forward_cosine": 0.9998434468319541,
+ "mean_feedback_forward_relative_error": 0.01735522203441513,
+ "min_feedback_forward_cosine": 0.9997166395187378
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 57086,
+ "train_examples": 50000,
+ "train_loss": 0.026103150560855864
+ },
+ {
+ "epoch": 147,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023825665935873985,
+ "mean_feedback_forward_cosine": 0.9998435629041571,
+ "mean_feedback_forward_relative_error": 0.017347723637756548,
+ "min_feedback_forward_cosine": 0.9997166395187378
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 57477,
+ "train_examples": 50000,
+ "train_loss": 0.026399266733825207
+ },
+ {
+ "epoch": 148,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02381902001798153,
+ "mean_feedback_forward_cosine": 0.9998436977988795,
+ "mean_feedback_forward_relative_error": 0.017339963318878097,
+ "min_feedback_forward_cosine": 0.9997169375419617
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 57868,
+ "train_examples": 50000,
+ "train_loss": 0.027654942645430565
+ },
+ {
+ "epoch": 149,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02381226420402527,
+ "mean_feedback_forward_cosine": 0.9998438044598228,
+ "mean_feedback_forward_relative_error": 0.01733238909295515,
+ "min_feedback_forward_cosine": 0.999717116355896
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 58259,
+ "train_examples": 50000,
+ "train_loss": 0.0263747742831707
+ },
+ {
+ "epoch": 150,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023805851116776466,
+ "mean_feedback_forward_cosine": 0.9998439079836795,
+ "mean_feedback_forward_relative_error": 0.017324834718908135,
+ "min_feedback_forward_cosine": 0.999717116355896
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 58650,
+ "train_examples": 50000,
+ "train_loss": 0.025472247238755227
+ },
+ {
+ "epoch": 151,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023805778473615646,
+ "mean_feedback_forward_cosine": 0.999843933080372,
+ "mean_feedback_forward_relative_error": 0.017324644091882203,
+ "min_feedback_forward_cosine": 0.9997171759605408
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 59041,
+ "train_examples": 50000,
+ "train_loss": 0.022873886232972145
+ },
+ {
+ "epoch": 152,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023805730044841766,
+ "mean_feedback_forward_cosine": 0.999843955039978,
+ "mean_feedback_forward_relative_error": 0.01732448201724573,
+ "min_feedback_forward_cosine": 0.9997171759605408
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 59432,
+ "train_examples": 50000,
+ "train_loss": 0.021361280524134635
+ },
+ {
+ "epoch": 153,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023805681616067886,
+ "mean_feedback_forward_cosine": 0.9998439519028914,
+ "mean_feedback_forward_relative_error": 0.01732430888928081,
+ "min_feedback_forward_cosine": 0.9997171759605408
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 59823,
+ "train_examples": 50000,
+ "train_loss": 0.020693761838823557
+ },
+ {
+ "epoch": 154,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023805633187294006,
+ "mean_feedback_forward_cosine": 0.9998439268061989,
+ "mean_feedback_forward_relative_error": 0.01732414387362568,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 60214,
+ "train_examples": 50000,
+ "train_loss": 0.021524387866258623
+ },
+ {
+ "epoch": 155,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023805582895874977,
+ "mean_feedback_forward_cosine": 0.9998439236691123,
+ "mean_feedback_forward_relative_error": 0.017323974912103853,
+ "min_feedback_forward_cosine": 0.9997171759605408
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 60605,
+ "train_examples": 50000,
+ "train_loss": 0.020588710283078255
+ },
+ {
+ "epoch": 156,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023805536329746246,
+ "mean_feedback_forward_cosine": 0.9998439393545452,
+ "mean_feedback_forward_relative_error": 0.01732380712698949,
+ "min_feedback_forward_cosine": 0.999717116355896
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 60996,
+ "train_examples": 50000,
+ "train_loss": 0.019683998126387596
+ },
+ {
+ "epoch": 157,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023805495351552963,
+ "mean_feedback_forward_cosine": 0.9998439519028914,
+ "mean_feedback_forward_relative_error": 0.017323645370963373,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 61387,
+ "train_examples": 50000,
+ "train_loss": 0.018572507655620574
+ },
+ {
+ "epoch": 158,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023805446922779083,
+ "mean_feedback_forward_cosine": 0.9998439487658048,
+ "mean_feedback_forward_relative_error": 0.017323482781648636,
+ "min_feedback_forward_cosine": 0.999717116355896
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 61778,
+ "train_examples": 50000,
+ "train_loss": 0.019311912797093392
+ },
+ {
+ "epoch": 159,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0238054022192955,
+ "mean_feedback_forward_cosine": 0.9998439393545452,
+ "mean_feedback_forward_relative_error": 0.01732332879481347,
+ "min_feedback_forward_cosine": 0.9997171759605408
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 62169,
+ "train_examples": 50000,
+ "train_loss": 0.019017863205373287
+ },
+ {
+ "epoch": 160,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02380535565316677,
+ "mean_feedback_forward_cosine": 0.9998439362174586,
+ "mean_feedback_forward_relative_error": 0.01732316380366683,
+ "min_feedback_forward_cosine": 0.999717116355896
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 62560,
+ "train_examples": 50000,
+ "train_loss": 0.01927336893439293
+ },
+ {
+ "epoch": 161,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02380530722439289,
+ "mean_feedback_forward_cosine": 0.9998439393545452,
+ "mean_feedback_forward_relative_error": 0.017323002635844444,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 62951,
+ "train_examples": 50000,
+ "train_loss": 0.018690299270153046
+ },
+ {
+ "epoch": 162,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02380525879561901,
+ "mean_feedback_forward_cosine": 0.999843955039978,
+ "mean_feedback_forward_relative_error": 0.017322842399344632,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 63342,
+ "train_examples": 50000,
+ "train_loss": 0.018778735616505147
+ },
+ {
+ "epoch": 163,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02380521036684513,
+ "mean_feedback_forward_cosine": 0.9998439519028914,
+ "mean_feedback_forward_relative_error": 0.017322684932304055,
+ "min_feedback_forward_cosine": 0.9997171759605408
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 63733,
+ "train_examples": 50000,
+ "train_loss": 0.018716709740757943
+ },
+ {
+ "epoch": 164,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0238051638007164,
+ "mean_feedback_forward_cosine": 0.999843933080372,
+ "mean_feedback_forward_relative_error": 0.01732252025976777,
+ "min_feedback_forward_cosine": 0.999717116355896
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 64124,
+ "train_examples": 50000,
+ "train_loss": 0.018794280961900948
+ },
+ {
+ "epoch": 165,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02380511350929737,
+ "mean_feedback_forward_cosine": 0.9998439456287184,
+ "mean_feedback_forward_relative_error": 0.017322357842012456,
+ "min_feedback_forward_cosine": 0.9997171759605408
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 64515,
+ "train_examples": 50000,
+ "train_loss": 0.01923126378715038
+ },
+ {
+ "epoch": 166,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02380507066845894,
+ "mean_feedback_forward_cosine": 0.9998439393545452,
+ "mean_feedback_forward_relative_error": 0.017322201085718053,
+ "min_feedback_forward_cosine": 0.9997171759605408
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 64906,
+ "train_examples": 50000,
+ "train_loss": 0.019187729357481004
+ },
+ {
+ "epoch": 167,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023805027827620506,
+ "mean_feedback_forward_cosine": 0.999843955039978,
+ "mean_feedback_forward_relative_error": 0.017322045726407515,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 65297,
+ "train_examples": 50000,
+ "train_loss": 0.017014023590385913
+ },
+ {
+ "epoch": 168,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023804988712072372,
+ "mean_feedback_forward_cosine": 0.9998439613141512,
+ "mean_feedback_forward_relative_error": 0.01732188144600705,
+ "min_feedback_forward_cosine": 0.999717116355896
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 65688,
+ "train_examples": 50000,
+ "train_loss": 0.017929602970182895
+ },
+ {
+ "epoch": 169,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02380494214594364,
+ "mean_feedback_forward_cosine": 0.9998439581770646,
+ "mean_feedback_forward_relative_error": 0.017321729664935878,
+ "min_feedback_forward_cosine": 0.9997171759605408
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 66079,
+ "train_examples": 50000,
+ "train_loss": 0.017645584962666036
+ },
+ {
+ "epoch": 170,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02380489557981491,
+ "mean_feedback_forward_cosine": 0.9998439456287184,
+ "mean_feedback_forward_relative_error": 0.01732157080091144,
+ "min_feedback_forward_cosine": 0.9997171759605408
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 66470,
+ "train_examples": 50000,
+ "train_loss": 0.01727688707411289
+ },
+ {
+ "epoch": 171,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023804854601621628,
+ "mean_feedback_forward_cosine": 0.9998439707254109,
+ "mean_feedback_forward_relative_error": 0.017321413088785976,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 66861,
+ "train_examples": 50000,
+ "train_loss": 0.017035351138710974
+ },
+ {
+ "epoch": 172,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023804813623428345,
+ "mean_feedback_forward_cosine": 0.9998439801366705,
+ "mean_feedback_forward_relative_error": 0.01732124819567329,
+ "min_feedback_forward_cosine": 0.9997172951698303
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 67252,
+ "train_examples": 50000,
+ "train_loss": 0.017451484676599502
+ },
+ {
+ "epoch": 173,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023804765194654465,
+ "mean_feedback_forward_cosine": 0.999843955039978,
+ "mean_feedback_forward_relative_error": 0.017321086170053797,
+ "min_feedback_forward_cosine": 0.9997171759605408
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 67643,
+ "train_examples": 50000,
+ "train_loss": 0.017583800478577615
+ },
+ {
+ "epoch": 174,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023804711177945137,
+ "mean_feedback_forward_cosine": 0.9998439675883243,
+ "mean_feedback_forward_relative_error": 0.0173209248061635,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 68034,
+ "train_examples": 50000,
+ "train_loss": 0.018306848403811454
+ },
+ {
+ "epoch": 175,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02380465902388096,
+ "mean_feedback_forward_cosine": 0.9998439801366705,
+ "mean_feedback_forward_relative_error": 0.01732077581906005,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 68425,
+ "train_examples": 50000,
+ "train_loss": 0.016635816239118577
+ },
+ {
+ "epoch": 176,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023804618045687675,
+ "mean_feedback_forward_cosine": 0.9998439675883243,
+ "mean_feedback_forward_relative_error": 0.017320614847305574,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 68816,
+ "train_examples": 50000,
+ "train_loss": 0.017419615784883498
+ },
+ {
+ "epoch": 177,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02380458079278469,
+ "mean_feedback_forward_cosine": 0.9998439487658048,
+ "mean_feedback_forward_relative_error": 0.01732046534552386,
+ "min_feedback_forward_cosine": 0.9997171759605408
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 69207,
+ "train_examples": 50000,
+ "train_loss": 0.016998306522369385
+ },
+ {
+ "epoch": 178,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02380453795194626,
+ "mean_feedback_forward_cosine": 0.999843955039978,
+ "mean_feedback_forward_relative_error": 0.017320308074551195,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 69598,
+ "train_examples": 50000,
+ "train_loss": 0.016233026780486107
+ },
+ {
+ "epoch": 179,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023804493248462677,
+ "mean_feedback_forward_cosine": 0.9998439832737571,
+ "mean_feedback_forward_relative_error": 0.017320153131884963,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 69989,
+ "train_examples": 50000,
+ "train_loss": 0.017383653934001923
+ },
+ {
+ "epoch": 180,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023804450407624245,
+ "mean_feedback_forward_cosine": 0.9998439613141512,
+ "mean_feedback_forward_relative_error": 0.017320000689084594,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 70380,
+ "train_examples": 50000,
+ "train_loss": 0.0170416913112998
+ },
+ {
+ "epoch": 181,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023804405704140663,
+ "mean_feedback_forward_cosine": 0.9998439738624975,
+ "mean_feedback_forward_relative_error": 0.017319846089537207,
+ "min_feedback_forward_cosine": 0.9997171759605408
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 70771,
+ "train_examples": 50000,
+ "train_loss": 0.016445232709050177
+ },
+ {
+ "epoch": 182,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02380436658859253,
+ "mean_feedback_forward_cosine": 0.999843976999584,
+ "mean_feedback_forward_relative_error": 0.017319693891821725,
+ "min_feedback_forward_cosine": 0.9997171759605408
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 71162,
+ "train_examples": 50000,
+ "train_loss": 0.016828604788780212
+ },
+ {
+ "epoch": 183,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023804321885108948,
+ "mean_feedback_forward_cosine": 0.9998439832737571,
+ "mean_feedback_forward_relative_error": 0.017319535395424617,
+ "min_feedback_forward_cosine": 0.9997171759605408
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 71553,
+ "train_examples": 50000,
+ "train_loss": 0.016447176468521358
+ },
+ {
+ "epoch": 184,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023804284632205963,
+ "mean_feedback_forward_cosine": 0.9998439707254109,
+ "mean_feedback_forward_relative_error": 0.017319383589844955,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 71944,
+ "train_examples": 50000,
+ "train_loss": 0.0159349295052886
+ },
+ {
+ "epoch": 185,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02380424551665783,
+ "mean_feedback_forward_cosine": 0.9998439519028914,
+ "mean_feedback_forward_relative_error": 0.01731923038728143,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 72335,
+ "train_examples": 50000,
+ "train_loss": 0.01665976164340973
+ },
+ {
+ "epoch": 186,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023804206401109695,
+ "mean_feedback_forward_cosine": 0.9998439738624975,
+ "mean_feedback_forward_relative_error": 0.01731907419468227,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 72726,
+ "train_examples": 50000,
+ "train_loss": 0.01594864286452532
+ },
+ {
+ "epoch": 187,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023804159834980965,
+ "mean_feedback_forward_cosine": 0.9998439738624975,
+ "mean_feedback_forward_relative_error": 0.017318920158830128,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 73117,
+ "train_examples": 50000,
+ "train_loss": 0.016114003044962882
+ },
+ {
+ "epoch": 188,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02380412444472313,
+ "mean_feedback_forward_cosine": 0.9998439864108437,
+ "mean_feedback_forward_relative_error": 0.01731876867186082,
+ "min_feedback_forward_cosine": 0.9997172951698303
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 73508,
+ "train_examples": 50000,
+ "train_loss": 0.016495792008042336
+ },
+ {
+ "epoch": 189,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023804085329174995,
+ "mean_feedback_forward_cosine": 0.9998439926850168,
+ "mean_feedback_forward_relative_error": 0.017318619366147016,
+ "min_feedback_forward_cosine": 0.9997171759605408
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 73899,
+ "train_examples": 50000,
+ "train_loss": 0.015994502281844614
+ },
+ {
+ "epoch": 190,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023804040625691414,
+ "mean_feedback_forward_cosine": 0.99984399895919,
+ "mean_feedback_forward_relative_error": 0.01731846827131353,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 74290,
+ "train_examples": 50000,
+ "train_loss": 0.016001024417877197
+ },
+ {
+ "epoch": 191,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023804010823369026,
+ "mean_feedback_forward_cosine": 0.999843976999584,
+ "mean_feedback_forward_relative_error": 0.017318317078446086,
+ "min_feedback_forward_cosine": 0.999717116355896
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 74681,
+ "train_examples": 50000,
+ "train_loss": 0.015285830289721488
+ },
+ {
+ "epoch": 192,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02380397729575634,
+ "mean_feedback_forward_cosine": 0.9998439707254109,
+ "mean_feedback_forward_relative_error": 0.01731816600812109,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 75072,
+ "train_examples": 50000,
+ "train_loss": 0.015755094451606274
+ },
+ {
+ "epoch": 193,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023803941905498505,
+ "mean_feedback_forward_cosine": 0.9998439644512377,
+ "mean_feedback_forward_relative_error": 0.01731802113844376,
+ "min_feedback_forward_cosine": 0.9997171759605408
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 75463,
+ "train_examples": 50000,
+ "train_loss": 0.014800295288711787
+ },
+ {
+ "epoch": 194,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02380390465259552,
+ "mean_feedback_forward_cosine": 0.9998439864108437,
+ "mean_feedback_forward_relative_error": 0.01731786680848975,
+ "min_feedback_forward_cosine": 0.9997171759605408
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 75854,
+ "train_examples": 50000,
+ "train_loss": 0.01677374741256237
+ },
+ {
+ "epoch": 195,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023803865537047386,
+ "mean_feedback_forward_cosine": 0.9998439895479303,
+ "mean_feedback_forward_relative_error": 0.017317713409858316,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 76245,
+ "train_examples": 50000,
+ "train_loss": 0.015875785137712956
+ },
+ {
+ "epoch": 196,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023803822696208954,
+ "mean_feedback_forward_cosine": 0.9998439926850168,
+ "mean_feedback_forward_relative_error": 0.017317562976754027,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 76636,
+ "train_examples": 50000,
+ "train_loss": 0.016230050567686557
+ },
+ {
+ "epoch": 197,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023803774267435074,
+ "mean_feedback_forward_cosine": 0.9998439801366705,
+ "mean_feedback_forward_relative_error": 0.017317412788734624,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 77027,
+ "train_examples": 50000,
+ "train_loss": 0.01620101815342903
+ },
+ {
+ "epoch": 198,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023803729563951492,
+ "mean_feedback_forward_cosine": 0.9998439675883243,
+ "mean_feedback_forward_relative_error": 0.017317260296917276,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 77418,
+ "train_examples": 50000,
+ "train_loss": 0.01728534370690584
+ },
+ {
+ "epoch": 199,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02380368486046791,
+ "mean_feedback_forward_cosine": 0.9998439832737571,
+ "mean_feedback_forward_relative_error": 0.017317102486757857,
+ "min_feedback_forward_cosine": 0.9997172355651855
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 77809,
+ "train_examples": 50000,
+ "train_loss": 0.01559103421419859
+ },
+ {
+ "epoch": 200,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02380364201962948,
+ "mean_feedback_forward_cosine": 0.9998440146446228,
+ "mean_feedback_forward_relative_error": 0.017316952690874274,
+ "min_feedback_forward_cosine": 0.9997172951698303
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 78200,
+ "train_examples": 50000,
+ "train_loss": 0.015727360000908373
+ }
+ ],
+ "evaluation_protocol": {
+ "test_evaluations": 1,
+ "test_used_for_selection": false,
+ "validation_evaluations": 0
+ },
+ "final": {
+ "accuracy": 0.9158,
+ "epoch": 200,
+ "evaluation_split": "test",
+ "finite": true,
+ "loss": 0.37618404846191406,
+ "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": 10,
+ "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": 0.7086873054504395,
+ "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": 6220.81570649147,
+ "train_wall_s": 6218.6675980091095
+ },
+ "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
+ }
+}