summaryrefslogtreecommitdiff
path: root/results
diff options
context:
space:
mode:
authorYurenHao0426 <Blackhao0426@gmail.com>2026-07-22 16:04:42 -0500
committerYurenHao0426 <Blackhao0426@gmail.com>2026-07-22 16:04:52 -0500
commit2ef7f94291c35a284394bf0dd2f0c452c89847a0 (patch)
treec259bdafa0f40911ddc3b46fa436369912509edb /results
parenta402546b794bf69ab79bce7ed6365cbdd1e7d15a (diff)
results: pass reciprocal KP full gate
Diffstat (limited to 'results')
-rw-r--r--results/kp_full/kp.json3256
-rw-r--r--results/kp_full_gate.json34
2 files changed, 3290 insertions, 0 deletions
diff --git a/results/kp_full/kp.json b/results/kp_full/kp.json
new file mode 100644
index 0000000..bbf090f
--- /dev/null
+++ b/results/kp_full/kp.json
@@ -0,0 +1,3256 @@
+{
+ "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": "validation",
+ "learn_P": 0,
+ "loader_seed": 0,
+ "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,
+ "normalization": "batchnorm",
+ "nuisance_scale": 0.0,
+ "out": "results/kp_full/kp.json",
+ "output_lr": 0.1,
+ "pert_directions": 1,
+ "pert_every": 4,
+ "pert_sigma": 0.01,
+ "perturb_seed": 1000,
+ "predictor_warmup_steps": 0,
+ "residual_scale": null,
+ "seed": 0,
+ "split_seed": 2027,
+ "train_limit": 0,
+ "use_residual": 1,
+ "val_examples": 5000,
+ "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,
+ "ordinary_examples": 9000000,
+ "per_example_loss_terms": 0,
+ "perturbation_events": 0,
+ "perturbation_forward_examples": 0,
+ "predictor_warmup_examples": 0
+ },
+ "diagnostics": {
+ "early_third_mean": 0.9993971586227417,
+ "feedback_forward_cosine": [
+ 0.9996035695075989,
+ 0.999546229839325,
+ 0.9996310472488403,
+ 0.9995880126953125,
+ 0.9996770024299622,
+ 0.9996175765991211,
+ 0.9996256828308105,
+ 0.9997471570968628,
+ 0.9996682405471802,
+ 0.9995927214622498,
+ 0.9996819496154785,
+ 0.9995782971382141,
+ 0.9996360540390015,
+ 0.999785304069519,
+ 0.9997695088386536,
+ 0.999726414680481,
+ 0.9997502565383911,
+ 0.9994064569473267,
+ 0.9999741911888123
+ ],
+ "feedback_forward_norm_ratio": [
+ 0.9998817443847656,
+ 1.0025546550750732,
+ 0.9983091950416565,
+ 1.0026606321334839,
+ 0.9995179176330566,
+ 1.0004081726074219,
+ 0.9985063672065735,
+ 1.0004786252975464,
+ 0.9998459815979004,
+ 1.001400113105774,
+ 1.0002866983413696,
+ 1.001346468925476,
+ 0.9995779395103455,
+ 1.0001710653305054,
+ 1.0000461339950562,
+ 1.0010814666748047,
+ 1.0001813173294067,
+ 1.000972867012024,
+ 1.0038318634033203
+ ],
+ "feedback_forward_relative_error": [
+ 0.028156239539384842,
+ 0.030272938311100006,
+ 0.02719682641327381,
+ 0.0288621224462986,
+ 0.025414466857910156,
+ 0.027663787826895714,
+ 0.027380341663956642,
+ 0.02249538153409958,
+ 0.025756139308214188,
+ 0.028594130650162697,
+ 0.025227291509509087,
+ 0.029092716053128242,
+ 0.026978736743330956,
+ 0.020724760368466377,
+ 0.021473316475749016,
+ 0.023429762572050095,
+ 0.02234940230846405,
+ 0.034483253955841064,
+ 0.00815806444734335
+ ],
+ "innovation_negative_gradient_cosine": [
+ 0.9994229078292847,
+ 0.9993288516998291,
+ 0.9994176626205444,
+ 0.9993650317192078,
+ 0.9993991255760193,
+ 0.9994493722915649,
+ 0.9994412660598755,
+ 0.9995417594909668,
+ 0.9995058178901672,
+ 0.9994058012962341,
+ 0.9995360970497131,
+ 0.9994885921478271,
+ 0.9995619058609009,
+ 0.9996987581253052,
+ 0.9997124671936035,
+ 0.9997549057006836,
+ 0.9998419284820557,
+ 0.999946117401123,
+ 0.8377549648284912
+ ],
+ "mean_feedback_forward_cosine": 0.9996634564901653,
+ "mean_feedback_forward_relative_error": 0.025458404157114655,
+ "normalization_state": "training_batch_stats_without_running_update",
+ "raw_negative_gradient_cosine": [
+ 0.9994229078292847,
+ 0.9993288516998291,
+ 0.9994176626205444,
+ 0.9993650317192078,
+ 0.9993991255760193,
+ 0.9994493722915649,
+ 0.9994412660598755,
+ 0.9995417594909668,
+ 0.9995058178901672,
+ 0.9994058012962341,
+ 0.9995360970497131,
+ 0.9994885921478271,
+ 0.9995619058609009,
+ 0.9996987581253052,
+ 0.9997124671936035,
+ 0.9997549057006836,
+ 0.9998419284820557,
+ 0.999946117401123,
+ 0.8377549648284912
+ ],
+ "teaching_negative_gradient_cosine": [
+ 0.9994229078292847,
+ 0.9993288516998291,
+ 0.9994176626205444,
+ 0.9993650317192078,
+ 0.9993991255760193,
+ 0.9994493722915649,
+ 0.9994412660598755,
+ 0.9995417594909668,
+ 0.9995058178901672,
+ 0.9994058012962341,
+ 0.9995360970497131,
+ 0.9994885921478271,
+ 0.9995619058609009,
+ 0.9996987581253052,
+ 0.9997124671936035,
+ 0.9997549057006836,
+ 0.9998419284820557,
+ 0.999946117401123,
+ 0.8377549648284912
+ ],
+ "wall_s": 0.1015174388885498
+ },
+ "epochs": [
+ {
+ "epoch": 1,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.3187716007232666,
+ "mean_feedback_forward_cosine": 0.2785400740410152,
+ "mean_feedback_forward_relative_error": 1.200114883874592,
+ "min_feedback_forward_cosine": 0.13789714872837067
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 352,
+ "train_examples": 45000,
+ "train_loss": 1.8397478539996677
+ },
+ {
+ "epoch": 2,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.2449522018432617,
+ "mean_feedback_forward_cosine": 0.4070323721358651,
+ "mean_feedback_forward_relative_error": 1.08963417065771,
+ "min_feedback_forward_cosine": 0.23210661113262177
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 704,
+ "train_examples": 45000,
+ "train_loss": 1.3883055899302166
+ },
+ {
+ "epoch": 3,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.1890921592712402,
+ "mean_feedback_forward_cosine": 0.499870617138712,
+ "mean_feedback_forward_relative_error": 1.0024236346545972,
+ "min_feedback_forward_cosine": 0.30416029691696167
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 1056,
+ "train_examples": 45000,
+ "train_loss": 1.1106628182093303
+ },
+ {
+ "epoch": 4,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.1485661268234253,
+ "mean_feedback_forward_cosine": 0.5687334662989566,
+ "mean_feedback_forward_relative_error": 0.9321056933779466,
+ "min_feedback_forward_cosine": 0.3554292321205139
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 1408,
+ "train_examples": 45000,
+ "train_loss": 0.9362988779385885
+ },
+ {
+ "epoch": 5,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.1077516078948975,
+ "mean_feedback_forward_cosine": 0.6216514549757305,
+ "mean_feedback_forward_relative_error": 0.8729972933468065,
+ "min_feedback_forward_cosine": 0.40342050790786743
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 1760,
+ "train_examples": 45000,
+ "train_loss": 0.8056027103000217
+ },
+ {
+ "epoch": 6,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.0688467025756836,
+ "mean_feedback_forward_cosine": 0.6630745937949732,
+ "mean_feedback_forward_relative_error": 0.8240033262654355,
+ "min_feedback_forward_cosine": 0.4471142292022705
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 2112,
+ "train_examples": 45000,
+ "train_loss": 0.720929954316881
+ },
+ {
+ "epoch": 7,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.031362771987915,
+ "mean_feedback_forward_cosine": 0.6980409669248682,
+ "mean_feedback_forward_relative_error": 0.7803165363638025,
+ "min_feedback_forward_cosine": 0.48720064759254456
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 2464,
+ "train_examples": 45000,
+ "train_loss": 0.6616328886773851
+ },
+ {
+ "epoch": 8,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.9942635893821716,
+ "mean_feedback_forward_cosine": 0.7275104240367287,
+ "mean_feedback_forward_relative_error": 0.7413108270419272,
+ "min_feedback_forward_cosine": 0.5240331888198853
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 2816,
+ "train_examples": 45000,
+ "train_loss": 0.6209926544507345
+ },
+ {
+ "epoch": 9,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.9602717757225037,
+ "mean_feedback_forward_cosine": 0.753035444962351,
+ "mean_feedback_forward_relative_error": 0.7060555643156955,
+ "min_feedback_forward_cosine": 0.5570434331893921
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 3168,
+ "train_examples": 45000,
+ "train_loss": 0.5786632180213929
+ },
+ {
+ "epoch": 10,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.9283965826034546,
+ "mean_feedback_forward_cosine": 0.774936716807516,
+ "mean_feedback_forward_relative_error": 0.6740853410018118,
+ "min_feedback_forward_cosine": 0.5865967273712158
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 3520,
+ "train_examples": 45000,
+ "train_loss": 0.5443803879949781
+ },
+ {
+ "epoch": 11,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.8959965705871582,
+ "mean_feedback_forward_cosine": 0.7947383428874769,
+ "mean_feedback_forward_relative_error": 0.6437877824431971,
+ "min_feedback_forward_cosine": 0.6155344247817993
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 3872,
+ "train_examples": 45000,
+ "train_loss": 0.520760157696406
+ },
+ {
+ "epoch": 12,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.8660936951637268,
+ "mean_feedback_forward_cosine": 0.8118437999173215,
+ "mean_feedback_forward_relative_error": 0.6161490757214395,
+ "min_feedback_forward_cosine": 0.6414409279823303
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 4224,
+ "train_examples": 45000,
+ "train_loss": 0.49678643630345665
+ },
+ {
+ "epoch": 13,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.8367525935173035,
+ "mean_feedback_forward_cosine": 0.8271583506935521,
+ "mean_feedback_forward_relative_error": 0.5902244574145267,
+ "min_feedback_forward_cosine": 0.6656643152236938
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 4576,
+ "train_examples": 45000,
+ "train_loss": 0.4738122732374403
+ },
+ {
+ "epoch": 14,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.8078418970108032,
+ "mean_feedback_forward_cosine": 0.8413542697304174,
+ "mean_feedback_forward_relative_error": 0.5653508838854338,
+ "min_feedback_forward_cosine": 0.6884337663650513
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 4928,
+ "train_examples": 45000,
+ "train_loss": 0.4631085410647922
+ },
+ {
+ "epoch": 15,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.7801172733306885,
+ "mean_feedback_forward_cosine": 0.853670459044607,
+ "mean_feedback_forward_relative_error": 0.5428898844279741,
+ "min_feedback_forward_cosine": 0.7093034982681274
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 5280,
+ "train_examples": 45000,
+ "train_loss": 0.4412741024653117
+ },
+ {
+ "epoch": 16,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.7536543607711792,
+ "mean_feedback_forward_cosine": 0.8648707647072641,
+ "mean_feedback_forward_relative_error": 0.5214008179150129,
+ "min_feedback_forward_cosine": 0.7286971807479858
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 5632,
+ "train_examples": 45000,
+ "train_loss": 0.42227579946517946
+ },
+ {
+ "epoch": 17,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.7269087433815002,
+ "mean_feedback_forward_cosine": 0.8751571711740995,
+ "mean_feedback_forward_relative_error": 0.5008949488401413,
+ "min_feedback_forward_cosine": 0.7475579977035522
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 5984,
+ "train_examples": 45000,
+ "train_loss": 0.41152598244349164
+ },
+ {
+ "epoch": 18,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.7006407976150513,
+ "mean_feedback_forward_cosine": 0.8847063905314395,
+ "mean_feedback_forward_relative_error": 0.4812463435687517,
+ "min_feedback_forward_cosine": 0.7653176784515381
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 6336,
+ "train_examples": 45000,
+ "train_loss": 0.40556067257457307
+ },
+ {
+ "epoch": 19,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.6762637495994568,
+ "mean_feedback_forward_cosine": 0.8930537104606628,
+ "mean_feedback_forward_relative_error": 0.46315100710642965,
+ "min_feedback_forward_cosine": 0.7813417911529541
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 6688,
+ "train_examples": 45000,
+ "train_loss": 0.3921797104358673
+ },
+ {
+ "epoch": 20,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.6523281335830688,
+ "mean_feedback_forward_cosine": 0.9008636568721972,
+ "mean_feedback_forward_relative_error": 0.44579507802662094,
+ "min_feedback_forward_cosine": 0.7964432835578918
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 7040,
+ "train_examples": 45000,
+ "train_loss": 0.3793167280726963
+ },
+ {
+ "epoch": 21,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.6290809512138367,
+ "mean_feedback_forward_cosine": 0.9080616863150346,
+ "mean_feedback_forward_relative_error": 0.4291515334656364,
+ "min_feedback_forward_cosine": 0.8104970455169678
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 7392,
+ "train_examples": 45000,
+ "train_loss": 0.3700509198506673
+ },
+ {
+ "epoch": 22,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.6061670184135437,
+ "mean_feedback_forward_cosine": 0.91462894176182,
+ "mean_feedback_forward_relative_error": 0.4133475916950326,
+ "min_feedback_forward_cosine": 0.8239156007766724
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 7744,
+ "train_examples": 45000,
+ "train_loss": 0.36224497916433546
+ },
+ {
+ "epoch": 23,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.5838472247123718,
+ "mean_feedback_forward_cosine": 0.9209115097397252,
+ "mean_feedback_forward_relative_error": 0.39759282218782527,
+ "min_feedback_forward_cosine": 0.8365517854690552
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 8096,
+ "train_examples": 45000,
+ "train_loss": 0.3537719224505954
+ },
+ {
+ "epoch": 24,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.5628406405448914,
+ "mean_feedback_forward_cosine": 0.9265054119260687,
+ "mean_feedback_forward_relative_error": 0.3830666008748506,
+ "min_feedback_forward_cosine": 0.8479509949684143
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 8448,
+ "train_examples": 45000,
+ "train_loss": 0.3417613176451789
+ },
+ {
+ "epoch": 25,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.5425907373428345,
+ "mean_feedback_forward_cosine": 0.9317423356206793,
+ "mean_feedback_forward_relative_error": 0.36904128993812363,
+ "min_feedback_forward_cosine": 0.858492374420166
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 8800,
+ "train_examples": 45000,
+ "train_loss": 0.33725900309350754
+ },
+ {
+ "epoch": 26,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.5235796570777893,
+ "mean_feedback_forward_cosine": 0.9365188761761314,
+ "mean_feedback_forward_relative_error": 0.3556948351232629,
+ "min_feedback_forward_cosine": 0.8681777715682983
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 9152,
+ "train_examples": 45000,
+ "train_loss": 0.33141013379096984
+ },
+ {
+ "epoch": 27,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.5044535994529724,
+ "mean_feedback_forward_cosine": 0.9410122005563033,
+ "mean_feedback_forward_relative_error": 0.3426245756839451,
+ "min_feedback_forward_cosine": 0.8774411678314209
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 9504,
+ "train_examples": 45000,
+ "train_loss": 0.32462858259942795
+ },
+ {
+ "epoch": 28,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.48616400361061096,
+ "mean_feedback_forward_cosine": 0.94509651472694,
+ "mean_feedback_forward_relative_error": 0.3303080083508241,
+ "min_feedback_forward_cosine": 0.8860575556755066
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 9856,
+ "train_examples": 45000,
+ "train_loss": 0.3190594089905421
+ },
+ {
+ "epoch": 29,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.4686717689037323,
+ "mean_feedback_forward_cosine": 0.9489053613261172,
+ "mean_feedback_forward_relative_error": 0.31855456295766327,
+ "min_feedback_forward_cosine": 0.8939859867095947
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 10208,
+ "train_examples": 45000,
+ "train_loss": 0.3160331652985679
+ },
+ {
+ "epoch": 30,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.45222729444503784,
+ "mean_feedback_forward_cosine": 0.95233448555595,
+ "mean_feedback_forward_relative_error": 0.30750202740493576,
+ "min_feedback_forward_cosine": 0.9010915756225586
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 10560,
+ "train_examples": 45000,
+ "train_loss": 0.3038366813871596
+ },
+ {
+ "epoch": 31,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.4352026581764221,
+ "mean_feedback_forward_cosine": 0.9556411222407692,
+ "mean_feedback_forward_relative_error": 0.296561042729177,
+ "min_feedback_forward_cosine": 0.9083324670791626
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 10912,
+ "train_examples": 45000,
+ "train_loss": 0.3052713762601217
+ },
+ {
+ "epoch": 32,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.41834455728530884,
+ "mean_feedback_forward_cosine": 0.958824634552002,
+ "mean_feedback_forward_relative_error": 0.28559827922206177,
+ "min_feedback_forward_cosine": 0.9151769876480103
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 11264,
+ "train_examples": 45000,
+ "train_loss": 0.3022956451786889
+ },
+ {
+ "epoch": 33,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.40313923358917236,
+ "mean_feedback_forward_cosine": 0.9616001781664396,
+ "mean_feedback_forward_relative_error": 0.27568476803992925,
+ "min_feedback_forward_cosine": 0.9212157130241394
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 11616,
+ "train_examples": 45000,
+ "train_loss": 0.2964079156557719
+ },
+ {
+ "epoch": 34,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.388301283121109,
+ "mean_feedback_forward_cosine": 0.9642708991703234,
+ "mean_feedback_forward_relative_error": 0.2657922514175114,
+ "min_feedback_forward_cosine": 0.9267534017562866
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 11968,
+ "train_examples": 45000,
+ "train_loss": 0.29282834168010285
+ },
+ {
+ "epoch": 35,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.374147891998291,
+ "mean_feedback_forward_cosine": 0.9667425312493977,
+ "mean_feedback_forward_relative_error": 0.256254894953025,
+ "min_feedback_forward_cosine": 0.9319241046905518
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 12320,
+ "train_examples": 45000,
+ "train_loss": 0.285447880686654
+ },
+ {
+ "epoch": 36,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.3606281876564026,
+ "mean_feedback_forward_cosine": 0.9690211885853818,
+ "mean_feedback_forward_relative_error": 0.2472790985515243,
+ "min_feedback_forward_cosine": 0.9366850256919861
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 12672,
+ "train_examples": 45000,
+ "train_loss": 0.2839041856235928
+ },
+ {
+ "epoch": 37,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.3475404381752014,
+ "mean_feedback_forward_cosine": 0.9711891757814508,
+ "mean_feedback_forward_relative_error": 0.23836818729576312,
+ "min_feedback_forward_cosine": 0.9411553144454956
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 13024,
+ "train_examples": 45000,
+ "train_loss": 0.2837774367438422
+ },
+ {
+ "epoch": 38,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.33517706394195557,
+ "mean_feedback_forward_cosine": 0.9731261134147644,
+ "mean_feedback_forward_relative_error": 0.23015287557714864,
+ "min_feedback_forward_cosine": 0.9452270269393921
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 13376,
+ "train_examples": 45000,
+ "train_loss": 0.2742618181652493
+ },
+ {
+ "epoch": 39,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.3227942883968353,
+ "mean_feedback_forward_cosine": 0.9749487011056197,
+ "mean_feedback_forward_relative_error": 0.22212200062839607,
+ "min_feedback_forward_cosine": 0.9491633176803589
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 13728,
+ "train_examples": 45000,
+ "train_loss": 0.26789241547584536
+ },
+ {
+ "epoch": 40,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.31103014945983887,
+ "mean_feedback_forward_cosine": 0.9766529955362019,
+ "mean_feedback_forward_relative_error": 0.21434897772575678,
+ "min_feedback_forward_cosine": 0.9527660608291626
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 14080,
+ "train_examples": 45000,
+ "train_loss": 0.2670468606524997
+ },
+ {
+ "epoch": 41,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.2992304861545563,
+ "mean_feedback_forward_cosine": 0.9782717635757044,
+ "mean_feedback_forward_relative_error": 0.2067208595966038,
+ "min_feedback_forward_cosine": 0.956220269203186
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 14432,
+ "train_examples": 45000,
+ "train_loss": 0.2691164943854014
+ },
+ {
+ "epoch": 42,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.2882028818130493,
+ "mean_feedback_forward_cosine": 0.9797702400307906,
+ "mean_feedback_forward_relative_error": 0.19939432606885307,
+ "min_feedback_forward_cosine": 0.9593504667282104
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 14784,
+ "train_examples": 45000,
+ "train_loss": 0.259995029815038
+ },
+ {
+ "epoch": 43,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.2778346836566925,
+ "mean_feedback_forward_cosine": 0.9811753034591675,
+ "mean_feedback_forward_relative_error": 0.1923066271763099,
+ "min_feedback_forward_cosine": 0.9621995687484741
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 15136,
+ "train_examples": 45000,
+ "train_loss": 0.2582001879665587
+ },
+ {
+ "epoch": 44,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.267698734998703,
+ "mean_feedback_forward_cosine": 0.9824458893976713,
+ "mean_feedback_forward_relative_error": 0.1856212882619155,
+ "min_feedback_forward_cosine": 0.9648654460906982
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 15488,
+ "train_examples": 45000,
+ "train_loss": 0.25434536108440825
+ },
+ {
+ "epoch": 45,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.25788941979408264,
+ "mean_feedback_forward_cosine": 0.98361715517546,
+ "mean_feedback_forward_relative_error": 0.1792655998938962,
+ "min_feedback_forward_cosine": 0.9673755168914795
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 15840,
+ "train_examples": 45000,
+ "train_loss": 0.25353836012416414
+ },
+ {
+ "epoch": 46,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.24856674671173096,
+ "mean_feedback_forward_cosine": 0.984717425547148,
+ "mean_feedback_forward_relative_error": 0.17313916824365916,
+ "min_feedback_forward_cosine": 0.9696618318557739
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 16192,
+ "train_examples": 45000,
+ "train_loss": 0.25382505049175685
+ },
+ {
+ "epoch": 47,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.23936481773853302,
+ "mean_feedback_forward_cosine": 0.9857612879652726,
+ "mean_feedback_forward_relative_error": 0.1670643448045379,
+ "min_feedback_forward_cosine": 0.9718400835990906
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 16544,
+ "train_examples": 45000,
+ "train_loss": 0.2500472250090705
+ },
+ {
+ "epoch": 48,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.23048098385334015,
+ "mean_feedback_forward_cosine": 0.9867121263554222,
+ "mean_feedback_forward_relative_error": 0.16135098863589137,
+ "min_feedback_forward_cosine": 0.9738771915435791
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 16896,
+ "train_examples": 45000,
+ "train_loss": 0.2480532353507148
+ },
+ {
+ "epoch": 49,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.22193792462348938,
+ "mean_feedback_forward_cosine": 0.9876200399900738,
+ "mean_feedback_forward_relative_error": 0.15570892903365588,
+ "min_feedback_forward_cosine": 0.9757627248764038
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 17248,
+ "train_examples": 45000,
+ "train_loss": 0.24684581592877705
+ },
+ {
+ "epoch": 50,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.21395453810691833,
+ "mean_feedback_forward_cosine": 0.9884608513430545,
+ "mean_feedback_forward_relative_error": 0.15029135660121315,
+ "min_feedback_forward_cosine": 0.9774685502052307
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 17600,
+ "train_examples": 45000,
+ "train_loss": 0.2417957257270813
+ },
+ {
+ "epoch": 51,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.206168994307518,
+ "mean_feedback_forward_cosine": 0.9892476702991285,
+ "mean_feedback_forward_relative_error": 0.1450631094998435,
+ "min_feedback_forward_cosine": 0.9790802001953125
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 17952,
+ "train_examples": 45000,
+ "train_loss": 0.24198342045678034
+ },
+ {
+ "epoch": 52,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.19910243153572083,
+ "mean_feedback_forward_cosine": 0.98995315087469,
+ "mean_feedback_forward_relative_error": 0.14018461363095985,
+ "min_feedback_forward_cosine": 0.9804839491844177
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 18304,
+ "train_examples": 45000,
+ "train_loss": 0.2335544925265842
+ },
+ {
+ "epoch": 53,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.19207394123077393,
+ "mean_feedback_forward_cosine": 0.9906234427502281,
+ "mean_feedback_forward_relative_error": 0.13540399505903847,
+ "min_feedback_forward_cosine": 0.9818247556686401
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 18656,
+ "train_examples": 45000,
+ "train_loss": 0.23128648313946193
+ },
+ {
+ "epoch": 54,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.18490546941757202,
+ "mean_feedback_forward_cosine": 0.991272371066244,
+ "mean_feedback_forward_relative_error": 0.13063357456734306,
+ "min_feedback_forward_cosine": 0.9831516742706299
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 19008,
+ "train_examples": 45000,
+ "train_loss": 0.23452207955784268
+ },
+ {
+ "epoch": 55,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.17812126874923706,
+ "mean_feedback_forward_cosine": 0.9918646781068099,
+ "mean_feedback_forward_relative_error": 0.12609902573259255,
+ "min_feedback_forward_cosine": 0.9843542575836182
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 19360,
+ "train_examples": 45000,
+ "train_loss": 0.23310066242218017
+ },
+ {
+ "epoch": 56,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.171734020113945,
+ "mean_feedback_forward_cosine": 0.9924089187069943,
+ "mean_feedback_forward_relative_error": 0.12177852345140357,
+ "min_feedback_forward_cosine": 0.9854464530944824
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 19712,
+ "train_examples": 45000,
+ "train_loss": 0.23404440263112386
+ },
+ {
+ "epoch": 57,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.16528268158435822,
+ "mean_feedback_forward_cosine": 0.9929253302122417,
+ "mean_feedback_forward_relative_error": 0.11753287283997786,
+ "min_feedback_forward_cosine": 0.9865061044692993
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 20064,
+ "train_examples": 45000,
+ "train_loss": 0.2302730461968316
+ },
+ {
+ "epoch": 58,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.1593792885541916,
+ "mean_feedback_forward_cosine": 0.993399723580009,
+ "mean_feedback_forward_relative_error": 0.11350665084625546,
+ "min_feedback_forward_cosine": 0.9874453544616699
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 20416,
+ "train_examples": 45000,
+ "train_loss": 0.22987807891633774
+ },
+ {
+ "epoch": 59,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.15377117693424225,
+ "mean_feedback_forward_cosine": 0.9938396686001828,
+ "mean_feedback_forward_relative_error": 0.10962733724399616,
+ "min_feedback_forward_cosine": 0.9883084297180176
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 20768,
+ "train_examples": 45000,
+ "train_loss": 0.22291153036753336
+ },
+ {
+ "epoch": 60,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.14817892014980316,
+ "mean_feedback_forward_cosine": 0.9942594327424702,
+ "mean_feedback_forward_relative_error": 0.1058036700675362,
+ "min_feedback_forward_cosine": 0.9891355037689209
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 21120,
+ "train_examples": 45000,
+ "train_loss": 0.22237795277701483
+ },
+ {
+ "epoch": 61,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.14284008741378784,
+ "mean_feedback_forward_cosine": 0.9946550946486624,
+ "mean_feedback_forward_relative_error": 0.10208968249590773,
+ "min_feedback_forward_cosine": 0.9899026155471802
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 21472,
+ "train_examples": 45000,
+ "train_loss": 0.224736327846845
+ },
+ {
+ "epoch": 62,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.1376606822013855,
+ "mean_feedback_forward_cosine": 0.9950216977219832,
+ "mean_feedback_forward_relative_error": 0.09851729222818424,
+ "min_feedback_forward_cosine": 0.9906183481216431
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 21824,
+ "train_examples": 45000,
+ "train_loss": 0.22513049019707573
+ },
+ {
+ "epoch": 63,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.13297568261623383,
+ "mean_feedback_forward_cosine": 0.9953495107199016,
+ "mean_feedback_forward_relative_error": 0.09519688216479201,
+ "min_feedback_forward_cosine": 0.9912426471710205
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 22176,
+ "train_examples": 45000,
+ "train_loss": 0.216873992840449
+ },
+ {
+ "epoch": 64,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.12823887169361115,
+ "mean_feedback_forward_cosine": 0.9956639973740828,
+ "mean_feedback_forward_relative_error": 0.09191947979362387,
+ "min_feedback_forward_cosine": 0.9918533563613892
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 22528,
+ "train_examples": 45000,
+ "train_loss": 0.21968923865954082
+ },
+ {
+ "epoch": 65,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.12350643426179886,
+ "mean_feedback_forward_cosine": 0.9959614810190702,
+ "mean_feedback_forward_relative_error": 0.08870825210684224,
+ "min_feedback_forward_cosine": 0.9924424290657043
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 22880,
+ "train_examples": 45000,
+ "train_loss": 0.22088376659552256
+ },
+ {
+ "epoch": 66,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.11920788139104843,
+ "mean_feedback_forward_cosine": 0.9962275592904342,
+ "mean_feedback_forward_relative_error": 0.08571936601870939,
+ "min_feedback_forward_cosine": 0.9929566383361816
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 23232,
+ "train_examples": 45000,
+ "train_loss": 0.2136418219036526
+ },
+ {
+ "epoch": 67,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.1148824393749237,
+ "mean_feedback_forward_cosine": 0.9964875202429923,
+ "mean_feedback_forward_relative_error": 0.0827051838369746,
+ "min_feedback_forward_cosine": 0.9934556484222412
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 23584,
+ "train_examples": 45000,
+ "train_loss": 0.2167039133389791
+ },
+ {
+ "epoch": 68,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.11087697744369507,
+ "mean_feedback_forward_cosine": 0.9967204771543804,
+ "mean_feedback_forward_relative_error": 0.07989444916969851,
+ "min_feedback_forward_cosine": 0.993901252746582
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 23936,
+ "train_examples": 45000,
+ "train_loss": 0.207972959862815
+ },
+ {
+ "epoch": 69,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.10684216767549515,
+ "mean_feedback_forward_cosine": 0.9969419993852314,
+ "mean_feedback_forward_relative_error": 0.07713936700632698,
+ "min_feedback_forward_cosine": 0.9943339824676514
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 24288,
+ "train_examples": 45000,
+ "train_loss": 0.21679914739131928
+ },
+ {
+ "epoch": 70,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.10296562314033508,
+ "mean_feedback_forward_cosine": 0.99715466875779,
+ "mean_feedback_forward_relative_error": 0.07441286772097412,
+ "min_feedback_forward_cosine": 0.9947364926338196
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 24640,
+ "train_examples": 45000,
+ "train_loss": 0.2153888538704978
+ },
+ {
+ "epoch": 71,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.09926202893257141,
+ "mean_feedback_forward_cosine": 0.9973512856583846,
+ "mean_feedback_forward_relative_error": 0.07178653502150585,
+ "min_feedback_forward_cosine": 0.9951065182685852
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 24992,
+ "train_examples": 45000,
+ "train_loss": 0.21278025257322522
+ },
+ {
+ "epoch": 72,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0958256796002388,
+ "mean_feedback_forward_cosine": 0.9975294470787048,
+ "mean_feedback_forward_relative_error": 0.06932672936665385,
+ "min_feedback_forward_cosine": 0.9954378008842468
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 25344,
+ "train_examples": 45000,
+ "train_loss": 0.20551657772196663
+ },
+ {
+ "epoch": 73,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0924200713634491,
+ "mean_feedback_forward_cosine": 0.9976969204450908,
+ "mean_feedback_forward_relative_error": 0.06692772790005333,
+ "min_feedback_forward_cosine": 0.995755136013031
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 25696,
+ "train_examples": 45000,
+ "train_loss": 0.2072883184168074
+ },
+ {
+ "epoch": 74,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.08904393017292023,
+ "mean_feedback_forward_cosine": 0.99785589230688,
+ "mean_feedback_forward_relative_error": 0.06457923960528876,
+ "min_feedback_forward_cosine": 0.9960581064224243
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 26048,
+ "train_examples": 45000,
+ "train_loss": 0.20923989634778764
+ },
+ {
+ "epoch": 75,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.08596943318843842,
+ "mean_feedback_forward_cosine": 0.9979973058951529,
+ "mean_feedback_forward_relative_error": 0.06240164478750605,
+ "min_feedback_forward_cosine": 0.9963244199752808
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 26400,
+ "train_examples": 45000,
+ "train_loss": 0.20445450791252984
+ },
+ {
+ "epoch": 76,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0829305574297905,
+ "mean_feedback_forward_cosine": 0.9981320845453363,
+ "mean_feedback_forward_relative_error": 0.06025716848671436,
+ "min_feedback_forward_cosine": 0.9965785145759583
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 26752,
+ "train_examples": 45000,
+ "train_loss": 0.20366583447986178
+ },
+ {
+ "epoch": 77,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.07978767156600952,
+ "mean_feedback_forward_cosine": 0.9982615395596153,
+ "mean_feedback_forward_relative_error": 0.0581331848117866,
+ "min_feedback_forward_cosine": 0.9968324303627014
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 27104,
+ "train_examples": 45000,
+ "train_loss": 0.20921306936475967
+ },
+ {
+ "epoch": 78,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.07686465233564377,
+ "mean_feedback_forward_cosine": 0.9983839486774645,
+ "mean_feedback_forward_relative_error": 0.05604745222157553,
+ "min_feedback_forward_cosine": 0.9970594644546509
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 27456,
+ "train_examples": 45000,
+ "train_loss": 0.20827081781493292
+ },
+ {
+ "epoch": 79,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0742088258266449,
+ "mean_feedback_forward_cosine": 0.9984873407765439,
+ "mean_feedback_forward_relative_error": 0.05421718268802291,
+ "min_feedback_forward_cosine": 0.9972594380378723
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 27808,
+ "train_examples": 45000,
+ "train_loss": 0.19697656389872234
+ },
+ {
+ "epoch": 80,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.07164964079856873,
+ "mean_feedback_forward_cosine": 0.9985895627423337,
+ "mean_feedback_forward_relative_error": 0.05235146878189162,
+ "min_feedback_forward_cosine": 0.9974445104598999
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 28160,
+ "train_examples": 45000,
+ "train_loss": 0.19833368481000266
+ },
+ {
+ "epoch": 81,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.06903065741062164,
+ "mean_feedback_forward_cosine": 0.9986871794650429,
+ "mean_feedback_forward_relative_error": 0.050506196622001495,
+ "min_feedback_forward_cosine": 0.9976270198822021
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 28512,
+ "train_examples": 45000,
+ "train_loss": 0.20593708272510106
+ },
+ {
+ "epoch": 82,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.066603884100914,
+ "mean_feedback_forward_cosine": 0.9987752061141165,
+ "mean_feedback_forward_relative_error": 0.04878160779021288,
+ "min_feedback_forward_cosine": 0.9977902173995972
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 28864,
+ "train_examples": 45000,
+ "train_loss": 0.20040477652549743
+ },
+ {
+ "epoch": 83,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.06430792808532715,
+ "mean_feedback_forward_cosine": 0.998856804872814,
+ "mean_feedback_forward_relative_error": 0.04712375390686487,
+ "min_feedback_forward_cosine": 0.9979394674301147
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 29216,
+ "train_examples": 45000,
+ "train_loss": 0.19574761364724902
+ },
+ {
+ "epoch": 84,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.06205420941114426,
+ "mean_feedback_forward_cosine": 0.9989350030296728,
+ "mean_feedback_forward_relative_error": 0.045481442914981594,
+ "min_feedback_forward_cosine": 0.9980815649032593
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 29568,
+ "train_examples": 45000,
+ "train_loss": 0.19576913906733195
+ },
+ {
+ "epoch": 85,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.05980977416038513,
+ "mean_feedback_forward_cosine": 0.9990088657328957,
+ "mean_feedback_forward_relative_error": 0.04387636149400159,
+ "min_feedback_forward_cosine": 0.9982175230979919
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 29920,
+ "train_examples": 45000,
+ "train_loss": 0.20140785335964626
+ },
+ {
+ "epoch": 86,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.057750146836042404,
+ "mean_feedback_forward_cosine": 0.9990746222044292,
+ "mean_feedback_forward_relative_error": 0.042391228832696616,
+ "min_feedback_forward_cosine": 0.9983378648757935
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 30272,
+ "train_examples": 45000,
+ "train_loss": 0.19004953569836086
+ },
+ {
+ "epoch": 87,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.05566445738077164,
+ "mean_feedback_forward_cosine": 0.9991381764411926,
+ "mean_feedback_forward_relative_error": 0.04091344822786356,
+ "min_feedback_forward_cosine": 0.998455822467804
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 30624,
+ "train_examples": 45000,
+ "train_loss": 0.20463992954889934
+ },
+ {
+ "epoch": 88,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.053801532834768295,
+ "mean_feedback_forward_cosine": 0.9991939632516158,
+ "mean_feedback_forward_relative_error": 0.039558475543009605,
+ "min_feedback_forward_cosine": 0.9985572099685669
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 30976,
+ "train_examples": 45000,
+ "train_loss": 0.19115953441725836
+ },
+ {
+ "epoch": 89,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.05185016244649887,
+ "mean_feedback_forward_cosine": 0.9992505406078539,
+ "mean_feedback_forward_relative_error": 0.038145716547181734,
+ "min_feedback_forward_cosine": 0.9986598491668701
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 31328,
+ "train_examples": 45000,
+ "train_loss": 0.19612533348666297
+ },
+ {
+ "epoch": 90,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.049983251839876175,
+ "mean_feedback_forward_cosine": 0.9993017127639369,
+ "mean_feedback_forward_relative_error": 0.03681598378247336,
+ "min_feedback_forward_cosine": 0.9987547397613525
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 31680,
+ "train_examples": 45000,
+ "train_loss": 0.19237803222603267
+ },
+ {
+ "epoch": 91,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.04813147708773613,
+ "mean_feedback_forward_cosine": 0.9993512912800437,
+ "mean_feedback_forward_relative_error": 0.0354862505766122,
+ "min_feedback_forward_cosine": 0.9988449215888977
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 32032,
+ "train_examples": 45000,
+ "train_loss": 0.19978077992598214
+ },
+ {
+ "epoch": 92,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.04635227099061012,
+ "mean_feedback_forward_cosine": 0.9993960167232313,
+ "mean_feedback_forward_relative_error": 0.03424219953778543,
+ "min_feedback_forward_cosine": 0.9989286065101624
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 32384,
+ "train_examples": 45000,
+ "train_loss": 0.19687706090609233
+ },
+ {
+ "epoch": 93,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.04470311105251312,
+ "mean_feedback_forward_cosine": 0.9994375486122934,
+ "mean_feedback_forward_relative_error": 0.033044266759565004,
+ "min_feedback_forward_cosine": 0.999003529548645
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 32736,
+ "train_examples": 45000,
+ "train_loss": 0.1959826779683431
+ },
+ {
+ "epoch": 94,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.04307663440704346,
+ "mean_feedback_forward_cosine": 0.9994755230451885,
+ "mean_feedback_forward_relative_error": 0.03190501754809367,
+ "min_feedback_forward_cosine": 0.9990745782852173
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 33088,
+ "train_examples": 45000,
+ "train_loss": 0.1984396495130327
+ },
+ {
+ "epoch": 95,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.04155827686190605,
+ "mean_feedback_forward_cosine": 0.9995106772372597,
+ "mean_feedback_forward_relative_error": 0.030818379190015167,
+ "min_feedback_forward_cosine": 0.999138355255127
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 33440,
+ "train_examples": 45000,
+ "train_loss": 0.19280920056766934
+ },
+ {
+ "epoch": 96,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.04013899341225624,
+ "mean_feedback_forward_cosine": 0.999542700616937,
+ "mean_feedback_forward_relative_error": 0.029787863604724407,
+ "min_feedback_forward_cosine": 0.9991960525512695
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 33792,
+ "train_examples": 45000,
+ "train_loss": 0.18911483534706963
+ },
+ {
+ "epoch": 97,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03867710381746292,
+ "mean_feedback_forward_cosine": 0.9995744793038619,
+ "mean_feedback_forward_relative_error": 0.02873471202818971,
+ "min_feedback_forward_cosine": 0.999253511428833
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 34144,
+ "train_examples": 45000,
+ "train_loss": 0.19185938526524438
+ },
+ {
+ "epoch": 98,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.037349406629800797,
+ "mean_feedback_forward_cosine": 0.9996032118797302,
+ "mean_feedback_forward_relative_error": 0.027745857344646203,
+ "min_feedback_forward_cosine": 0.9993038177490234
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 34496,
+ "train_examples": 45000,
+ "train_loss": 0.18631468467182583
+ },
+ {
+ "epoch": 99,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.036000389605760574,
+ "mean_feedback_forward_cosine": 0.9996311664581299,
+ "mean_feedback_forward_relative_error": 0.026753132525635392,
+ "min_feedback_forward_cosine": 0.9993530511856079
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 34848,
+ "train_examples": 45000,
+ "train_loss": 0.19572784990469613
+ },
+ {
+ "epoch": 100,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03477846458554268,
+ "mean_feedback_forward_cosine": 0.999655240460446,
+ "mean_feedback_forward_relative_error": 0.025862311111076883,
+ "min_feedback_forward_cosine": 0.9993963241577148
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 35200,
+ "train_examples": 45000,
+ "train_loss": 0.1849142056332694
+ },
+ {
+ "epoch": 101,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03476789966225624,
+ "mean_feedback_forward_cosine": 0.9996555196611505,
+ "mean_feedback_forward_relative_error": 0.025849742716864535,
+ "min_feedback_forward_cosine": 0.9993966221809387
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 35552,
+ "train_examples": 45000,
+ "train_loss": 0.11685399395757251
+ },
+ {
+ "epoch": 102,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03476205840706825,
+ "mean_feedback_forward_cosine": 0.9996556545558729,
+ "mean_feedback_forward_relative_error": 0.025840042992249915,
+ "min_feedback_forward_cosine": 0.9993968605995178
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 35904,
+ "train_examples": 45000,
+ "train_loss": 0.09023945558865865
+ },
+ {
+ "epoch": 103,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0347563773393631,
+ "mean_feedback_forward_cosine": 0.9996558302327206,
+ "mean_feedback_forward_relative_error": 0.02583088676788305,
+ "min_feedback_forward_cosine": 0.9993970394134521
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 36256,
+ "train_examples": 45000,
+ "train_loss": 0.08250192235575782
+ },
+ {
+ "epoch": 104,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03475096821784973,
+ "mean_feedback_forward_cosine": 0.9996559839499625,
+ "mean_feedback_forward_relative_error": 0.025822071750697336,
+ "min_feedback_forward_cosine": 0.9993972182273865
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 36608,
+ "train_examples": 45000,
+ "train_loss": 0.07450271632406447
+ },
+ {
+ "epoch": 105,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03474564105272293,
+ "mean_feedback_forward_cosine": 0.9996560937479922,
+ "mean_feedback_forward_relative_error": 0.025813605783409195,
+ "min_feedback_forward_cosine": 0.9993973970413208
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 36960,
+ "train_examples": 45000,
+ "train_loss": 0.0699736884435018
+ },
+ {
+ "epoch": 106,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03474041074514389,
+ "mean_feedback_forward_cosine": 0.9996563039327923,
+ "mean_feedback_forward_relative_error": 0.025805333188097728,
+ "min_feedback_forward_cosine": 0.9993976354598999
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 37312,
+ "train_examples": 45000,
+ "train_loss": 0.06487165032360288
+ },
+ {
+ "epoch": 107,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03473511338233948,
+ "mean_feedback_forward_cosine": 0.999656407456649,
+ "mean_feedback_forward_relative_error": 0.02579726210158122,
+ "min_feedback_forward_cosine": 0.9993977546691895
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 37664,
+ "train_examples": 45000,
+ "train_loss": 0.061982938067118326
+ },
+ {
+ "epoch": 108,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034729812294244766,
+ "mean_feedback_forward_cosine": 0.9996565580368042,
+ "mean_feedback_forward_relative_error": 0.025789191211132628,
+ "min_feedback_forward_cosine": 0.9993979334831238
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 38016,
+ "train_examples": 45000,
+ "train_loss": 0.059803799949420824
+ },
+ {
+ "epoch": 109,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03472437337040901,
+ "mean_feedback_forward_cosine": 0.9996566992056998,
+ "mean_feedback_forward_relative_error": 0.02578137315025455,
+ "min_feedback_forward_cosine": 0.9993981719017029
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 38368,
+ "train_examples": 45000,
+ "train_loss": 0.058093480832046934
+ },
+ {
+ "epoch": 110,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034719087183475494,
+ "mean_feedback_forward_cosine": 0.9996568121408161,
+ "mean_feedback_forward_relative_error": 0.025773888355807253,
+ "min_feedback_forward_cosine": 0.9993983507156372
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 38720,
+ "train_examples": 45000,
+ "train_loss": 0.05606751591099633
+ },
+ {
+ "epoch": 111,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03471382334828377,
+ "mean_feedback_forward_cosine": 0.9996569752693176,
+ "mean_feedback_forward_relative_error": 0.025766217296845036,
+ "min_feedback_forward_cosine": 0.9993985891342163
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 39072,
+ "train_examples": 45000,
+ "train_loss": 0.051685227584838865
+ },
+ {
+ "epoch": 112,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03470851108431816,
+ "mean_feedback_forward_cosine": 0.9996571446719923,
+ "mean_feedback_forward_relative_error": 0.025758614028362853,
+ "min_feedback_forward_cosine": 0.9993987083435059
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 39424,
+ "train_examples": 45000,
+ "train_loss": 0.05137180427445306
+ },
+ {
+ "epoch": 113,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0347030945122242,
+ "mean_feedback_forward_cosine": 0.9996572732925415,
+ "mean_feedback_forward_relative_error": 0.025751124087132905,
+ "min_feedback_forward_cosine": 0.999398946762085
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 39776,
+ "train_examples": 45000,
+ "train_loss": 0.04984426314300961
+ },
+ {
+ "epoch": 114,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0346977673470974,
+ "mean_feedback_forward_cosine": 0.9996574270097833,
+ "mean_feedback_forward_relative_error": 0.025743580129193634,
+ "min_feedback_forward_cosine": 0.999398946762085
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 40128,
+ "train_examples": 45000,
+ "train_loss": 0.04824412440922525
+ },
+ {
+ "epoch": 115,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03469261899590492,
+ "mean_feedback_forward_cosine": 0.9996575901382848,
+ "mean_feedback_forward_relative_error": 0.025736133764056784,
+ "min_feedback_forward_cosine": 0.9993993043899536
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 40480,
+ "train_examples": 45000,
+ "train_loss": 0.044838647851513494
+ },
+ {
+ "epoch": 116,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0346871018409729,
+ "mean_feedback_forward_cosine": 0.9996577156217474,
+ "mean_feedback_forward_relative_error": 0.02572857912041639,
+ "min_feedback_forward_cosine": 0.9993994235992432
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 40832,
+ "train_examples": 45000,
+ "train_loss": 0.04527742183473375
+ },
+ {
+ "epoch": 117,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03468160331249237,
+ "mean_feedback_forward_cosine": 0.999657856790643,
+ "mean_feedback_forward_relative_error": 0.02572123583798346,
+ "min_feedback_forward_cosine": 0.9993996620178223
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 41184,
+ "train_examples": 45000,
+ "train_loss": 0.04366959934234619
+ },
+ {
+ "epoch": 118,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034676067531108856,
+ "mean_feedback_forward_cosine": 0.9996580167820579,
+ "mean_feedback_forward_relative_error": 0.025713593159851274,
+ "min_feedback_forward_cosine": 0.9993998408317566
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 41536,
+ "train_examples": 45000,
+ "train_loss": 0.04317836588819822
+ },
+ {
+ "epoch": 119,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034670427441596985,
+ "mean_feedback_forward_cosine": 0.9996581673622131,
+ "mean_feedback_forward_relative_error": 0.02570617703818961,
+ "min_feedback_forward_cosine": 0.9993999600410461
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 41888,
+ "train_examples": 45000,
+ "train_loss": 0.04118238543238905
+ },
+ {
+ "epoch": 120,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034664615988731384,
+ "mean_feedback_forward_cosine": 0.9996583242165414,
+ "mean_feedback_forward_relative_error": 0.02569874199597459,
+ "min_feedback_forward_cosine": 0.99940025806427
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 42240,
+ "train_examples": 45000,
+ "train_loss": 0.04075602948533164
+ },
+ {
+ "epoch": 121,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034659482538700104,
+ "mean_feedback_forward_cosine": 0.9996584559741774,
+ "mean_feedback_forward_relative_error": 0.025691540813759753,
+ "min_feedback_forward_cosine": 0.9994003772735596
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 42592,
+ "train_examples": 45000,
+ "train_loss": 0.03776900245845318
+ },
+ {
+ "epoch": 122,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0346539244055748,
+ "mean_feedback_forward_cosine": 0.9996586159655922,
+ "mean_feedback_forward_relative_error": 0.02568423468619585,
+ "min_feedback_forward_cosine": 0.9994006156921387
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 42944,
+ "train_examples": 45000,
+ "train_loss": 0.03804491649551524
+ },
+ {
+ "epoch": 123,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0346485860645771,
+ "mean_feedback_forward_cosine": 0.9996587728199205,
+ "mean_feedback_forward_relative_error": 0.025677015269665343,
+ "min_feedback_forward_cosine": 0.999400794506073
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 43296,
+ "train_examples": 45000,
+ "train_loss": 0.03678913566933738
+ },
+ {
+ "epoch": 124,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03464358299970627,
+ "mean_feedback_forward_cosine": 0.9996589045775565,
+ "mean_feedback_forward_relative_error": 0.02566992172873334,
+ "min_feedback_forward_cosine": 0.9994009733200073
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 43648,
+ "train_examples": 45000,
+ "train_loss": 0.034033451684316
+ },
+ {
+ "epoch": 125,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034638192504644394,
+ "mean_feedback_forward_cosine": 0.9996590457464519,
+ "mean_feedback_forward_relative_error": 0.02566258594589798,
+ "min_feedback_forward_cosine": 0.9994012117385864
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 44000,
+ "train_examples": 45000,
+ "train_loss": 0.0351742676400476
+ },
+ {
+ "epoch": 126,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03463255614042282,
+ "mean_feedback_forward_cosine": 0.9996592182862131,
+ "mean_feedback_forward_relative_error": 0.02565536265702624,
+ "min_feedback_forward_cosine": 0.9994012713432312
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 44352,
+ "train_examples": 45000,
+ "train_loss": 0.03448911868201362
+ },
+ {
+ "epoch": 127,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03462681546807289,
+ "mean_feedback_forward_cosine": 0.9996593657292818,
+ "mean_feedback_forward_relative_error": 0.025647955064318682,
+ "min_feedback_forward_cosine": 0.9994015693664551
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 44704,
+ "train_examples": 45000,
+ "train_loss": 0.03420423005885548
+ },
+ {
+ "epoch": 128,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03462139517068863,
+ "mean_feedback_forward_cosine": 0.999659516309437,
+ "mean_feedback_forward_relative_error": 0.025640798634604403,
+ "min_feedback_forward_cosine": 0.9994018077850342
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 45056,
+ "train_examples": 45000,
+ "train_loss": 0.031466926051510705
+ },
+ {
+ "epoch": 129,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034615613520145416,
+ "mean_feedback_forward_cosine": 0.9996596574783325,
+ "mean_feedback_forward_relative_error": 0.025633296782248897,
+ "min_feedback_forward_cosine": 0.9994019269943237
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 45408,
+ "train_examples": 45000,
+ "train_loss": 0.03245527929200066
+ },
+ {
+ "epoch": 130,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034609824419021606,
+ "mean_feedback_forward_cosine": 0.99965984256644,
+ "mean_feedback_forward_relative_error": 0.025625813556344884,
+ "min_feedback_forward_cosine": 0.9994021058082581
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 45760,
+ "train_examples": 45000,
+ "train_loss": 0.03059288759695159
+ },
+ {
+ "epoch": 131,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034603994339704514,
+ "mean_feedback_forward_cosine": 0.9996599994207683,
+ "mean_feedback_forward_relative_error": 0.025618238374590874,
+ "min_feedback_forward_cosine": 0.9994022846221924
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 46112,
+ "train_examples": 45000,
+ "train_loss": 0.030520318817098935
+ },
+ {
+ "epoch": 132,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03459795564413071,
+ "mean_feedback_forward_cosine": 0.9996601562750967,
+ "mean_feedback_forward_relative_error": 0.025610548248024362,
+ "min_feedback_forward_cosine": 0.9994024634361267
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 46464,
+ "train_examples": 45000,
+ "train_loss": 0.030415536801020306
+ },
+ {
+ "epoch": 133,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034592270851135254,
+ "mean_feedback_forward_cosine": 0.999660313129425,
+ "mean_feedback_forward_relative_error": 0.02560292400027576,
+ "min_feedback_forward_cosine": 0.9994027614593506
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 46816,
+ "train_examples": 45000,
+ "train_loss": 0.029513800981640817
+ },
+ {
+ "epoch": 134,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03458646312355995,
+ "mean_feedback_forward_cosine": 0.9996604605724937,
+ "mean_feedback_forward_relative_error": 0.0255955238581488,
+ "min_feedback_forward_cosine": 0.9994028210639954
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 47168,
+ "train_examples": 45000,
+ "train_loss": 0.02869330367181036
+ },
+ {
+ "epoch": 135,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034580204635858536,
+ "mean_feedback_forward_cosine": 0.9996606801685534,
+ "mean_feedback_forward_relative_error": 0.02558756820661457,
+ "min_feedback_forward_cosine": 0.9994031190872192
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 47520,
+ "train_examples": 45000,
+ "train_loss": 0.030226403964890375
+ },
+ {
+ "epoch": 136,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03457477316260338,
+ "mean_feedback_forward_cosine": 0.9996607993778429,
+ "mean_feedback_forward_relative_error": 0.025580379082576224,
+ "min_feedback_forward_cosine": 0.9994032979011536
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 47872,
+ "train_examples": 45000,
+ "train_loss": 0.02586192560725742
+ },
+ {
+ "epoch": 137,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0345681793987751,
+ "mean_feedback_forward_cosine": 0.9996609907401236,
+ "mean_feedback_forward_relative_error": 0.025572237019476137,
+ "min_feedback_forward_cosine": 0.9994035959243774
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 48224,
+ "train_examples": 45000,
+ "train_loss": 0.02932392889145348
+ },
+ {
+ "epoch": 138,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034562040120363235,
+ "mean_feedback_forward_cosine": 0.9996611883765772,
+ "mean_feedback_forward_relative_error": 0.02556428974984508,
+ "min_feedback_forward_cosine": 0.9994038343429565
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 48576,
+ "train_examples": 45000,
+ "train_loss": 0.028187815578447448
+ },
+ {
+ "epoch": 139,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03455587849020958,
+ "mean_feedback_forward_cosine": 0.9996613358196459,
+ "mean_feedback_forward_relative_error": 0.025556357136290325,
+ "min_feedback_forward_cosine": 0.9994039535522461
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 48928,
+ "train_examples": 45000,
+ "train_loss": 0.026993994370102883
+ },
+ {
+ "epoch": 140,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03455015644431114,
+ "mean_feedback_forward_cosine": 0.9996615271819266,
+ "mean_feedback_forward_relative_error": 0.02554876788666374,
+ "min_feedback_forward_cosine": 0.9994043111801147
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 49280,
+ "train_examples": 45000,
+ "train_loss": 0.024927383714914323
+ },
+ {
+ "epoch": 141,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03454364836215973,
+ "mean_feedback_forward_cosine": 0.9996617028587743,
+ "mean_feedback_forward_relative_error": 0.025540753610824283,
+ "min_feedback_forward_cosine": 0.9994043707847595
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 49632,
+ "train_examples": 45000,
+ "train_loss": 0.026976652889119254
+ },
+ {
+ "epoch": 142,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03453773260116577,
+ "mean_feedback_forward_cosine": 0.9996618597131026,
+ "mean_feedback_forward_relative_error": 0.02553301368300852,
+ "min_feedback_forward_cosine": 0.9994047284126282
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 49984,
+ "train_examples": 45000,
+ "train_loss": 0.024446142086055545
+ },
+ {
+ "epoch": 143,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03453129529953003,
+ "mean_feedback_forward_cosine": 0.9996620636237296,
+ "mean_feedback_forward_relative_error": 0.0255249235779047,
+ "min_feedback_forward_cosine": 0.9994048476219177
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 50336,
+ "train_examples": 45000,
+ "train_loss": 0.025978597725762262
+ },
+ {
+ "epoch": 144,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034524962306022644,
+ "mean_feedback_forward_cosine": 0.9996622204780579,
+ "mean_feedback_forward_relative_error": 0.025517003708764127,
+ "min_feedback_forward_cosine": 0.9994050860404968
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 50688,
+ "train_examples": 45000,
+ "train_loss": 0.02413371960024039
+ },
+ {
+ "epoch": 145,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03451862931251526,
+ "mean_feedback_forward_cosine": 0.9996624181145116,
+ "mean_feedback_forward_relative_error": 0.025509139767995005,
+ "min_feedback_forward_cosine": 0.9994052052497864
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 51040,
+ "train_examples": 45000,
+ "train_loss": 0.024469557727707758
+ },
+ {
+ "epoch": 146,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0345119833946228,
+ "mean_feedback_forward_cosine": 0.999662603202619,
+ "mean_feedback_forward_relative_error": 0.02550077889310686,
+ "min_feedback_forward_cosine": 0.9994053840637207
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 51392,
+ "train_examples": 45000,
+ "train_loss": 0.025060760457317033
+ },
+ {
+ "epoch": 147,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03450515493750572,
+ "mean_feedback_forward_cosine": 0.9996628102503324,
+ "mean_feedback_forward_relative_error": 0.0254922135684051,
+ "min_feedback_forward_cosine": 0.9994058012962341
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 51744,
+ "train_examples": 45000,
+ "train_loss": 0.02573590273724662
+ },
+ {
+ "epoch": 148,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034498561173677444,
+ "mean_feedback_forward_cosine": 0.9996629890642668,
+ "mean_feedback_forward_relative_error": 0.025484196449580946,
+ "min_feedback_forward_cosine": 0.9994059801101685
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 52096,
+ "train_examples": 45000,
+ "train_loss": 0.02323672732412815
+ },
+ {
+ "epoch": 149,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034491851925849915,
+ "mean_feedback_forward_cosine": 0.9996631961119803,
+ "mean_feedback_forward_relative_error": 0.025475973606501754,
+ "min_feedback_forward_cosine": 0.9994063377380371
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 52448,
+ "train_examples": 45000,
+ "train_loss": 0.02313196766111586
+ },
+ {
+ "epoch": 150,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448551520705223,
+ "mean_feedback_forward_cosine": 0.9996633561033952,
+ "mean_feedback_forward_relative_error": 0.02546797820219868,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 52800,
+ "train_examples": 45000,
+ "train_loss": 0.021565465519328912
+ },
+ {
+ "epoch": 151,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448543697595596,
+ "mean_feedback_forward_cosine": 0.9996633435550489,
+ "mean_feedback_forward_relative_error": 0.025467751988847005,
+ "min_feedback_forward_cosine": 0.9994063377380371
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 53152,
+ "train_examples": 45000,
+ "train_loss": 0.020494685203333696
+ },
+ {
+ "epoch": 152,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448537364602089,
+ "mean_feedback_forward_cosine": 0.9996633686517414,
+ "mean_feedback_forward_relative_error": 0.02546755033300111,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 53504,
+ "train_examples": 45000,
+ "train_loss": 0.020441280234853428
+ },
+ {
+ "epoch": 153,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448532521724701,
+ "mean_feedback_forward_cosine": 0.9996633247325295,
+ "mean_feedback_forward_relative_error": 0.025467347353696823,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 53856,
+ "train_examples": 45000,
+ "train_loss": 0.018196273184484905
+ },
+ {
+ "epoch": 154,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448528051376343,
+ "mean_feedback_forward_cosine": 0.9996633561033952,
+ "mean_feedback_forward_relative_error": 0.02546714260978134,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 54208,
+ "train_examples": 45000,
+ "train_loss": 0.017958002071910433
+ },
+ {
+ "epoch": 155,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448523208498955,
+ "mean_feedback_forward_cosine": 0.9996633466921354,
+ "mean_feedback_forward_relative_error": 0.025466943453801304,
+ "min_feedback_forward_cosine": 0.9994063377380371
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 54560,
+ "train_examples": 45000,
+ "train_loss": 0.01828998305698236
+ },
+ {
+ "epoch": 156,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448517993092537,
+ "mean_feedback_forward_cosine": 0.9996633561033952,
+ "mean_feedback_forward_relative_error": 0.025466745964398508,
+ "min_feedback_forward_cosine": 0.9994062781333923
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 54912,
+ "train_examples": 45000,
+ "train_loss": 0.01763483314745956
+ },
+ {
+ "epoch": 157,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448512405157089,
+ "mean_feedback_forward_cosine": 0.9996633937484339,
+ "mean_feedback_forward_relative_error": 0.02546655166109926,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 55264,
+ "train_examples": 45000,
+ "train_loss": 0.019247104206681253
+ },
+ {
+ "epoch": 158,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448506072163582,
+ "mean_feedback_forward_cosine": 0.9996633780630011,
+ "mean_feedback_forward_relative_error": 0.02546635500498508,
+ "min_feedback_forward_cosine": 0.9994065165519714
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 55616,
+ "train_examples": 45000,
+ "train_loss": 0.017701718068081473
+ },
+ {
+ "epoch": 159,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034485019743442535,
+ "mean_feedback_forward_cosine": 0.9996633749259146,
+ "mean_feedback_forward_relative_error": 0.025466150555171464,
+ "min_feedback_forward_cosine": 0.9994063973426819
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 55968,
+ "train_examples": 45000,
+ "train_loss": 0.017024116742445364
+ },
+ {
+ "epoch": 160,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034484975039958954,
+ "mean_feedback_forward_cosine": 0.999663371788828,
+ "mean_feedback_forward_relative_error": 0.025465945713222027,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 56320,
+ "train_examples": 45000,
+ "train_loss": 0.017757106672103205
+ },
+ {
+ "epoch": 161,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034484922885894775,
+ "mean_feedback_forward_cosine": 0.9996633937484339,
+ "mean_feedback_forward_relative_error": 0.02546574743954759,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 56672,
+ "train_examples": 45000,
+ "train_loss": 0.017177313058243857
+ },
+ {
+ "epoch": 162,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448488563299179,
+ "mean_feedback_forward_cosine": 0.9996633812000877,
+ "mean_feedback_forward_relative_error": 0.025465559557472404,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 57024,
+ "train_examples": 45000,
+ "train_loss": 0.0153671736597187
+ },
+ {
+ "epoch": 163,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448483720421791,
+ "mean_feedback_forward_cosine": 0.9996633529663086,
+ "mean_feedback_forward_relative_error": 0.025465358097694422,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 57376,
+ "train_examples": 45000,
+ "train_loss": 0.016437109147508937
+ },
+ {
+ "epoch": 164,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034484799951314926,
+ "mean_feedback_forward_cosine": 0.9996633874742609,
+ "mean_feedback_forward_relative_error": 0.025465177127013083,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 57728,
+ "train_examples": 45000,
+ "train_loss": 0.015015156511465708
+ },
+ {
+ "epoch": 165,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448475897312164,
+ "mean_feedback_forward_cosine": 0.9996633749259146,
+ "mean_feedback_forward_relative_error": 0.02546498625490226,
+ "min_feedback_forward_cosine": 0.9994063973426819
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 58080,
+ "train_examples": 45000,
+ "train_loss": 0.016280568671557638
+ },
+ {
+ "epoch": 166,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448471799492836,
+ "mean_feedback_forward_cosine": 0.9996634062967802,
+ "mean_feedback_forward_relative_error": 0.02546478925566924,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 58432,
+ "train_examples": 45000,
+ "train_loss": 0.016651724920670193
+ },
+ {
+ "epoch": 167,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448467329144478,
+ "mean_feedback_forward_cosine": 0.9996634031596937,
+ "mean_feedback_forward_relative_error": 0.02546459500138697,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 58784,
+ "train_examples": 45000,
+ "train_loss": 0.01643339535974794
+ },
+ {
+ "epoch": 168,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448464348912239,
+ "mean_feedback_forward_cosine": 0.9996633874742609,
+ "mean_feedback_forward_relative_error": 0.02546440682520992,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 59136,
+ "train_examples": 45000,
+ "train_loss": 0.015089965336521467
+ },
+ {
+ "epoch": 169,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448459133505821,
+ "mean_feedback_forward_cosine": 0.9996633749259146,
+ "mean_feedback_forward_relative_error": 0.025464214188487905,
+ "min_feedback_forward_cosine": 0.9994063973426819
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 59488,
+ "train_examples": 45000,
+ "train_loss": 0.016565012350347307
+ },
+ {
+ "epoch": 170,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448455035686493,
+ "mean_feedback_forward_cosine": 0.99966341570804,
+ "mean_feedback_forward_relative_error": 0.02546402243407149,
+ "min_feedback_forward_cosine": 0.9994063973426819
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 59840,
+ "train_examples": 45000,
+ "train_loss": 0.01590943037337727
+ },
+ {
+ "epoch": 171,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448450565338135,
+ "mean_feedback_forward_cosine": 0.9996633937484339,
+ "mean_feedback_forward_relative_error": 0.025463827346500597,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 60192,
+ "train_examples": 45000,
+ "train_loss": 0.01631662884718842
+ },
+ {
+ "epoch": 172,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448445722460747,
+ "mean_feedback_forward_cosine": 0.9996633843371743,
+ "mean_feedback_forward_relative_error": 0.02546363711161049,
+ "min_feedback_forward_cosine": 0.9994063973426819
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 60544,
+ "train_examples": 45000,
+ "train_loss": 0.01647938998275333
+ },
+ {
+ "epoch": 173,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034484416246414185,
+ "mean_feedback_forward_cosine": 0.9996634031596937,
+ "mean_feedback_forward_relative_error": 0.02546344378865079,
+ "min_feedback_forward_cosine": 0.9994065165519714
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 60896,
+ "train_examples": 45000,
+ "train_loss": 0.01630753223763572
+ },
+ {
+ "epoch": 174,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0344843752682209,
+ "mean_feedback_forward_cosine": 0.9996634125709534,
+ "mean_feedback_forward_relative_error": 0.02546325580854165,
+ "min_feedback_forward_cosine": 0.9994065165519714
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 61248,
+ "train_examples": 45000,
+ "train_loss": 0.015515399742043681
+ },
+ {
+ "epoch": 175,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448432683944702,
+ "mean_feedback_forward_cosine": 0.99966341570804,
+ "mean_feedback_forward_relative_error": 0.025463061309174487,
+ "min_feedback_forward_cosine": 0.9994063973426819
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 61600,
+ "train_examples": 45000,
+ "train_loss": 0.01612479949808783
+ },
+ {
+ "epoch": 176,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034484293311834335,
+ "mean_feedback_forward_cosine": 0.9996634031596937,
+ "mean_feedback_forward_relative_error": 0.025462869456724116,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 61952,
+ "train_examples": 45000,
+ "train_loss": 0.014161571666184399
+ },
+ {
+ "epoch": 177,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448425233364105,
+ "mean_feedback_forward_cosine": 0.99966341570804,
+ "mean_feedback_forward_relative_error": 0.02546268368237897,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 62304,
+ "train_examples": 45000,
+ "train_loss": 0.015663991013997132
+ },
+ {
+ "epoch": 178,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034484200179576874,
+ "mean_feedback_forward_cosine": 0.9996634188451266,
+ "mean_feedback_forward_relative_error": 0.02546249202599651,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 62656,
+ "train_examples": 45000,
+ "train_loss": 0.01664738964041074
+ },
+ {
+ "epoch": 179,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448415920138359,
+ "mean_feedback_forward_cosine": 0.9996634251192996,
+ "mean_feedback_forward_relative_error": 0.025462297624663302,
+ "min_feedback_forward_cosine": 0.9994065761566162
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 63008,
+ "train_examples": 45000,
+ "train_loss": 0.014807501598364777
+ },
+ {
+ "epoch": 180,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448411822319031,
+ "mean_feedback_forward_cosine": 0.9996634125709534,
+ "mean_feedback_forward_relative_error": 0.025462112634589796,
+ "min_feedback_forward_cosine": 0.9994065761566162
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 63360,
+ "train_examples": 45000,
+ "train_loss": 0.014790897508131133
+ },
+ {
+ "epoch": 181,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034484077244997025,
+ "mean_feedback_forward_cosine": 0.99966341570804,
+ "mean_feedback_forward_relative_error": 0.025461930242416105,
+ "min_feedback_forward_cosine": 0.9994065165519714
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 63712,
+ "train_examples": 45000,
+ "train_loss": 0.014952515361706416
+ },
+ {
+ "epoch": 182,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448403254151344,
+ "mean_feedback_forward_cosine": 0.999663443941819,
+ "mean_feedback_forward_relative_error": 0.0254617480953273,
+ "min_feedback_forward_cosine": 0.9994065761566162
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 64064,
+ "train_examples": 45000,
+ "train_loss": 0.015111820505393876
+ },
+ {
+ "epoch": 183,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448398783802986,
+ "mean_feedback_forward_cosine": 0.999663421982213,
+ "mean_feedback_forward_relative_error": 0.02546156388952544,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 64416,
+ "train_examples": 45000,
+ "train_loss": 0.01586712514526314
+ },
+ {
+ "epoch": 184,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448394313454628,
+ "mean_feedback_forward_cosine": 0.9996634345305594,
+ "mean_feedback_forward_relative_error": 0.025461373507584397,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 64768,
+ "train_examples": 45000,
+ "train_loss": 0.016010563597745366
+ },
+ {
+ "epoch": 185,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0344838947057724,
+ "mean_feedback_forward_cosine": 0.9996634188451266,
+ "mean_feedback_forward_relative_error": 0.025461187439137382,
+ "min_feedback_forward_cosine": 0.9994065761566162
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 65120,
+ "train_examples": 45000,
+ "train_loss": 0.015971294322113198
+ },
+ {
+ "epoch": 186,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448385372757912,
+ "mean_feedback_forward_cosine": 0.9996634251192996,
+ "mean_feedback_forward_relative_error": 0.025461001909877125,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 65472,
+ "train_examples": 45000,
+ "train_loss": 0.014779627677632703
+ },
+ {
+ "epoch": 187,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034483812749385834,
+ "mean_feedback_forward_cosine": 0.999663421982213,
+ "mean_feedback_forward_relative_error": 0.025460818537363882,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 65824,
+ "train_examples": 45000,
+ "train_loss": 0.015454550893604755
+ },
+ {
+ "epoch": 188,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448377177119255,
+ "mean_feedback_forward_cosine": 0.9996634533530787,
+ "mean_feedback_forward_relative_error": 0.02546062830247377,
+ "min_feedback_forward_cosine": 0.9994065165519714
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 66176,
+ "train_examples": 45000,
+ "train_loss": 0.014300745720623268
+ },
+ {
+ "epoch": 189,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448372706770897,
+ "mean_feedback_forward_cosine": 0.9996634125709534,
+ "mean_feedback_forward_relative_error": 0.02546043894988926,
+ "min_feedback_forward_cosine": 0.9994063973426819
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 66528,
+ "train_examples": 45000,
+ "train_loss": 0.015399348079330391
+ },
+ {
+ "epoch": 190,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034483689814805984,
+ "mean_feedback_forward_cosine": 0.9996634282563862,
+ "mean_feedback_forward_relative_error": 0.025460253077510157,
+ "min_feedback_forward_cosine": 0.9994065165519714
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 66880,
+ "train_examples": 45000,
+ "train_loss": 0.014467445031139586
+ },
+ {
+ "epoch": 191,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034483641386032104,
+ "mean_feedback_forward_cosine": 0.9996634376676459,
+ "mean_feedback_forward_relative_error": 0.02546007078337042,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 67232,
+ "train_examples": 45000,
+ "train_loss": 0.014093352991342544
+ },
+ {
+ "epoch": 192,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448360040783882,
+ "mean_feedback_forward_cosine": 0.9996634313934728,
+ "mean_feedback_forward_relative_error": 0.025459887753976018,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 67584,
+ "train_examples": 45000,
+ "train_loss": 0.014737001321050856
+ },
+ {
+ "epoch": 193,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448355942964554,
+ "mean_feedback_forward_cosine": 0.9996634564901653,
+ "mean_feedback_forward_relative_error": 0.025459696783831243,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 67936,
+ "train_examples": 45000,
+ "train_loss": 0.01454441604581144
+ },
+ {
+ "epoch": 194,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034483522176742554,
+ "mean_feedback_forward_cosine": 0.9996634564901653,
+ "mean_feedback_forward_relative_error": 0.025459504147109232,
+ "min_feedback_forward_cosine": 0.9994065761566162
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 68288,
+ "train_examples": 45000,
+ "train_loss": 0.013707617290152444
+ },
+ {
+ "epoch": 195,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034483473747968674,
+ "mean_feedback_forward_cosine": 0.9996634408047325,
+ "mean_feedback_forward_relative_error": 0.025459322735275094,
+ "min_feedback_forward_cosine": 0.9994065761566162
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 68640,
+ "train_examples": 45000,
+ "train_loss": 0.014107471609446738
+ },
+ {
+ "epoch": 196,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448342904448509,
+ "mean_feedback_forward_cosine": 0.9996634564901653,
+ "mean_feedback_forward_relative_error": 0.025459141029339088,
+ "min_feedback_forward_cosine": 0.9994065761566162
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 68992,
+ "train_examples": 45000,
+ "train_loss": 0.015630016327401002
+ },
+ {
+ "epoch": 197,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448338434100151,
+ "mean_feedback_forward_cosine": 0.9996634470789056,
+ "mean_feedback_forward_relative_error": 0.025458956137299538,
+ "min_feedback_forward_cosine": 0.9994065165519714
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 69344,
+ "train_examples": 45000,
+ "train_loss": 0.014596812316278617
+ },
+ {
+ "epoch": 198,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448333591222763,
+ "mean_feedback_forward_cosine": 0.9996634690385116,
+ "mean_feedback_forward_relative_error": 0.025458775019567264,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 69696,
+ "train_examples": 45000,
+ "train_loss": 0.014847102849516604
+ },
+ {
+ "epoch": 199,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03448329120874405,
+ "mean_feedback_forward_cosine": 0.9996634502159921,
+ "mean_feedback_forward_relative_error": 0.025458594833157565,
+ "min_feedback_forward_cosine": 0.9994065165519714
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 70048,
+ "train_examples": 45000,
+ "train_loss": 0.013949833128021822
+ },
+ {
+ "epoch": 200,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.034483253955841064,
+ "mean_feedback_forward_cosine": 0.9996634564901653,
+ "mean_feedback_forward_relative_error": 0.025458404157114655,
+ "min_feedback_forward_cosine": 0.9994064569473267
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 70400,
+ "train_examples": 45000,
+ "train_loss": 0.014527444395754072
+ }
+ ],
+ "evaluation_protocol": {
+ "test_evaluations": 0,
+ "test_used_for_selection": false,
+ "validation_evaluations": 1
+ },
+ "final": {
+ "accuracy": 0.9126,
+ "epoch": 200,
+ "evaluation_split": "validation",
+ "finite": true,
+ "loss": 0.42901277465820314,
+ "step": 70400
+ },
+ "hardware": {
+ "cuda_device_name": "NVIDIA GeForce GTX 1080",
+ "cuda_visible_devices": "7",
+ "device": "cuda",
+ "device_total_memory_bytes": 8507949056,
+ "peak_memory_allocated_bytes": 2162790400,
+ "peak_memory_reserved_bytes": 2866806784,
+ "torch_version": "2.3.1+cu118"
+ },
+ "protocol_family": "oral_a_cifar_local_resnet_development",
+ "provenance": {
+ "git_commit": "44b110ffe144c06e7404b1fbd46c1a637638fe59",
+ "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": "validation",
+ "input_layout": "NCHW",
+ "input_shape": [
+ 3,
+ 32,
+ 32
+ ],
+ "loader_seed": 0,
+ "normalization_mean": [
+ 0.49140000343322754,
+ 0.4821999967098236,
+ 0.4465000033378601
+ ],
+ "normalization_std": [
+ 0.24699999392032623,
+ 0.2434999942779541,
+ 0.26159998774528503
+ ],
+ "split_from_training_only": true,
+ "split_seed": 2027,
+ "test_examples": 10000,
+ "train_examples": 45000,
+ "training_augmentation": "random_crop_32_padding_4_zero_then_horizontal_flip_p0.5",
+ "validation_class_counts": {
+ "0": 500,
+ "1": 500,
+ "2": 500,
+ "3": 500,
+ "4": 500,
+ "5": 500,
+ "6": 500,
+ "7": 500,
+ "8": 500,
+ "9": 500
+ },
+ "validation_examples": 5000,
+ "validation_index_sha256": "8328b206a97c420e49e54e3eca4abe3274c4756b084355784ea3fb8059e4515b"
+ },
+ "timing": {
+ "apical_warmup_wall_s": 0.0,
+ "evaluation_wall_s": 0.36924219131469727,
+ "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": 5579.69712305069,
+ "train_wall_s": 5577.880968332291
+ },
+ "work": {
+ "apical_macs_per_example": 40108672,
+ "causal_scalar_observations": 0,
+ "components": {
+ "apical_projection_macs": 360978048000000,
+ "apical_regression_macs": 0,
+ "bp_reverse_macs_estimate": 0,
+ "kp_reciprocal_correlation_macs": 360978048000000,
+ "local_weight_correlation_macs": 364959360000000,
+ "mirror_feedback_prediction_macs": 0,
+ "mirror_local_correlation_macs": 0,
+ "mirror_response_macs": 0,
+ "ordinary_forward_macs": 364959360000000,
+ "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; elementwise nonlinearities and optimizer arithmetic excluded",
+ "forward_macs_per_example": 40551040,
+ "logical_batch_loss_queries": 0,
+ "mirror_probe_examples": 0,
+ "mirror_readout_probe_examples": 0,
+ "per_example_cross_entropy_terms": 0,
+ "total_clean_forward_examples": 9000000,
+ "total_forward_equivalent_examples": 9000000,
+ "total_macs_estimate": 1451874816000000
+ }
+}
diff --git a/results/kp_full_gate.json b/results/kp_full_gate.json
new file mode 100644
index 0000000..90bf8ff
--- /dev/null
+++ b/results/kp_full_gate.json
@@ -0,0 +1,34 @@
+{
+ "checks": {
+ "accuracy_at_least_0.88": true,
+ "early_alignment_at_least_0.80": true,
+ "epoch151_to200_feedback_cosine_at_least_0.95": true,
+ "final_feedback_cosine_at_least_0.95": true,
+ "macs_at_most_1.40x_bp": true,
+ "record_and_trajectory_finite": true,
+ "zero_task_loss_queries": true
+ },
+ "confirmation_test_seeds_touched": false,
+ "innovation_experiment_opened": true,
+ "metrics": {
+ "accuracy": 0.9126,
+ "bp_accuracy": 0.9162,
+ "bp_total_macs": 1094878080000000,
+ "early_third_alignment": 0.9993971586227417,
+ "epoch151_to200_mean_feedback_forward_cosine": 0.9996634071124229,
+ "final_mean_feedback_forward_cosine": 0.9996634564901653,
+ "final_mean_feedback_forward_relative_error": 0.025458404157114655,
+ "logical_batch_loss_queries": 0,
+ "loss": 0.42901277465820314,
+ "mac_ratio_to_bp": 1.3260607208430002,
+ "peak_memory_allocated_bytes": 2162790400,
+ "source_commit": "44b110ffe144c06e7404b1fbd46c1a637638fe59",
+ "total_macs": 1451874816000000,
+ "wall_s": 5579.69712305069
+ },
+ "protocol": "kolen_pollack_full_v1",
+ "review_score_after": 5,
+ "review_score_before": 5,
+ "score_change_rule": "inherited KP baseline cannot raise score",
+ "status": "passed"
+}