summaryrefslogtreecommitdiff
path: root/results/kp_dynamic_projection_confirmation/seed14_clean_kp.json
diff options
context:
space:
mode:
Diffstat (limited to 'results/kp_dynamic_projection_confirmation/seed14_clean_kp.json')
-rw-r--r--results/kp_dynamic_projection_confirmation/seed14_clean_kp.json3258
1 files changed, 3258 insertions, 0 deletions
diff --git a/results/kp_dynamic_projection_confirmation/seed14_clean_kp.json b/results/kp_dynamic_projection_confirmation/seed14_clean_kp.json
new file mode 100644
index 0000000..3886c19
--- /dev/null
+++ b/results/kp_dynamic_projection_confirmation/seed14_clean_kp.json
@@ -0,0 +1,3258 @@
+{
+ "architecture": {
+ "adaptive_apical_parameters": 0,
+ "adaptive_feedback_parameters": 267904,
+ "base_width": 16,
+ "blocks_per_stage": 3,
+ "bn_eps": 1e-05,
+ "bn_momentum": 0.1,
+ "depth": 20,
+ "family": "CIFAR 6n+2 ResNet, option-A shortcuts",
+ "fixed_feedback_parameters": 0,
+ "fixed_traffic_coefficients": 0,
+ "forward_parameters": 269722,
+ "hidden_shapes": [
+ [
+ 16,
+ 32,
+ 32
+ ],
+ [
+ 16,
+ 32,
+ 32
+ ],
+ [
+ 16,
+ 32,
+ 32
+ ],
+ [
+ 16,
+ 32,
+ 32
+ ],
+ [
+ 16,
+ 32,
+ 32
+ ],
+ [
+ 16,
+ 32,
+ 32
+ ],
+ [
+ 16,
+ 32,
+ 32
+ ],
+ [
+ 32,
+ 16,
+ 16
+ ],
+ [
+ 32,
+ 16,
+ 16
+ ],
+ [
+ 32,
+ 16,
+ 16
+ ],
+ [
+ 32,
+ 16,
+ 16
+ ],
+ [
+ 32,
+ 16,
+ 16
+ ],
+ [
+ 32,
+ 16,
+ 16
+ ],
+ [
+ 64,
+ 8,
+ 8
+ ],
+ [
+ 64,
+ 8,
+ 8
+ ],
+ [
+ 64,
+ 8,
+ 8
+ ],
+ [
+ 64,
+ 8,
+ 8
+ ],
+ [
+ 64,
+ 8,
+ 8
+ ],
+ [
+ 64,
+ 8,
+ 8
+ ]
+ ],
+ "normalization": "batchnorm",
+ "predictor_parameters": 0,
+ "residual_scale": 1.0,
+ "vectorizer_mode": null,
+ "vectorizer_parameters": 0
+ },
+ "args": {
+ "a_scale": 1.0,
+ "a_warmup_steps": 0,
+ "alignment_probe": 32,
+ "apical_calibration_mode": "unit_targets",
+ "apical_seed": null,
+ "augment_train": 1,
+ "batch_size": 128,
+ "bn_eps": 1e-05,
+ "bn_momentum": 0.1,
+ "data_dir": "/home/yurenh2/sdrn/data",
+ "depth": 20,
+ "device": "cuda",
+ "epochs": 200,
+ "eta_A": 0.01,
+ "eta_P": 0.01,
+ "eval_every": 0,
+ "eval_split": "test",
+ "learn_P": 0,
+ "loader_seed": 14,
+ "lr": 0.1,
+ "lr_gamma": 0.1,
+ "lr_milestones": "100,150",
+ "lr_schedule": "step",
+ "max_steps": 0,
+ "mirror_batch_size": 1,
+ "mirror_eta": 0.1,
+ "mirror_every": 16,
+ "mirror_noise_std": 1.0,
+ "mirror_seed": 3000,
+ "mirror_warmup_steps": 0,
+ "mode": "kp",
+ "momentum": 0.9,
+ "neutral_projection": 0,
+ "normalization": "batchnorm",
+ "nuisance_scale": 0.0,
+ "out": "results/kp_dynamic_projection_confirmation/seed14_clean_kp.json",
+ "output_lr": 0.1,
+ "pert_directions": 1,
+ "pert_every": 4,
+ "pert_sigma": 0.01,
+ "perturb_seed": 1000,
+ "predictor_every": 0,
+ "predictor_mode": "nlms",
+ "predictor_warmup_steps": 0,
+ "residual_scale": null,
+ "seed": 14,
+ "split_seed": 2027,
+ "traffic_calibration_examples": 64,
+ "traffic_ratio": 4.0,
+ "traffic_rule": "innovation",
+ "traffic_seed": 4000,
+ "train_limit": 0,
+ "use_residual": 1,
+ "val_examples": 0,
+ "vectorizer_mode": "spatial_template",
+ "warmup_epochs": 0,
+ "weight_decay": 0.0001,
+ "weight_scale": 1.0,
+ "width": 16
+ },
+ "calibration_metric_space": "reciprocal_local_activity_products",
+ "counters": {
+ "apical_warmup_examples": 0,
+ "calibration_event_examples": 0,
+ "causal_scalar_observations": 0,
+ "logical_batch_loss_queries": 0,
+ "mirror_conv_examples": 0,
+ "mirror_events": 0,
+ "mirror_readout_examples": 0,
+ "neutral_projection_examples": 0,
+ "ordinary_examples": 10000000,
+ "per_example_loss_terms": 0,
+ "perturbation_events": 0,
+ "perturbation_forward_examples": 0,
+ "predictor_update_examples": 0,
+ "predictor_warmup_examples": 0,
+ "traffic_audit_examples": 0,
+ "traffic_calibration_examples": 0
+ },
+ "diagnostics": {
+ "early_third_mean": 0.9996606707572937,
+ "feedback_forward_cosine": [
+ 0.9998397827148438,
+ 0.9998446702957153,
+ 0.9997676610946655,
+ 0.9997360706329346,
+ 0.9998356103897095,
+ 0.9997872114181519,
+ 0.9998396039009094,
+ 0.9998887777328491,
+ 0.9998351335525513,
+ 0.9998185634613037,
+ 0.9998612403869629,
+ 0.9998190402984619,
+ 0.9998319149017334,
+ 0.9999021291732788,
+ 0.9998922348022461,
+ 0.9998776912689209,
+ 0.9998843669891357,
+ 0.9997295141220093,
+ 0.9999887943267822
+ ],
+ "feedback_forward_norm_ratio": [
+ 1.000428318977356,
+ 1.0008574724197388,
+ 0.9990053176879883,
+ 1.0012600421905518,
+ 0.9996617436408997,
+ 1.000982642173767,
+ 0.9993184208869934,
+ 1.0003552436828613,
+ 0.9997528791427612,
+ 1.0007574558258057,
+ 0.9996054172515869,
+ 1.0006641149520874,
+ 0.9997743368148804,
+ 1.0001587867736816,
+ 1.0000267028808594,
+ 1.0004032850265503,
+ 1.0001193284988403,
+ 1.000847339630127,
+ 1.0024750232696533
+ ],
+ "feedback_forward_relative_error": [
+ 0.017910555005073547,
+ 0.017654424533247948,
+ 0.021565964445471764,
+ 0.023022618144750595,
+ 0.018133867532014847,
+ 0.020668482407927513,
+ 0.017918230965733528,
+ 0.014916742220520973,
+ 0.01815802976489067,
+ 0.019074484705924988,
+ 0.016659196466207504,
+ 0.01903541572391987,
+ 0.018337950110435486,
+ 0.013992981053888798,
+ 0.014677700586616993,
+ 0.015658369287848473,
+ 0.01521098893135786,
+ 0.02328561618924141,
+ 0.005356569774448872
+ ],
+ "innovation_negative_gradient_cosine": [
+ 0.9996907711029053,
+ 0.9996677041053772,
+ 0.9996755719184875,
+ 0.9995771050453186,
+ 0.9996970891952515,
+ 0.9996557831764221,
+ 0.9997193217277527,
+ 0.9997643232345581,
+ 0.9997644424438477,
+ 0.9997292757034302,
+ 0.9997795820236206,
+ 0.9997588396072388,
+ 0.9997851848602295,
+ 0.9998385906219482,
+ 0.9998672008514404,
+ 0.9998890161514282,
+ 0.9999305009841919,
+ 0.9999741911888123,
+ 0.8421580791473389
+ ],
+ "mean_feedback_forward_cosine": 0.9998410532349035,
+ "mean_feedback_forward_relative_error": 0.01743358883418535,
+ "normalization_state": "training_batch_stats_without_running_update",
+ "raw_negative_gradient_cosine": [
+ 0.9996907711029053,
+ 0.9996677041053772,
+ 0.9996755719184875,
+ 0.9995771050453186,
+ 0.9996970891952515,
+ 0.9996557831764221,
+ 0.9997193217277527,
+ 0.9997643232345581,
+ 0.9997644424438477,
+ 0.9997292757034302,
+ 0.9997795820236206,
+ 0.9997588396072388,
+ 0.9997851848602295,
+ 0.9998385906219482,
+ 0.9998672008514404,
+ 0.9998890161514282,
+ 0.9999305009841919,
+ 0.9999741911888123,
+ 0.8421580791473389
+ ],
+ "teaching_negative_gradient_cosine": [
+ 0.9996907711029053,
+ 0.9996677041053772,
+ 0.9996755719184875,
+ 0.9995771050453186,
+ 0.9996970891952515,
+ 0.9996557831764221,
+ 0.9997193217277527,
+ 0.9997643232345581,
+ 0.9997644424438477,
+ 0.9997292757034302,
+ 0.9997795820236206,
+ 0.9997588396072388,
+ 0.9997851848602295,
+ 0.9998385906219482,
+ 0.9998672008514404,
+ 0.9998890161514282,
+ 0.9999305009841919,
+ 0.9999741911888123,
+ 0.8421580791473389
+ ],
+ "wall_s": 0.08478927612304688
+ },
+ "epochs": [
+ {
+ "epoch": 1,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.3152871131896973,
+ "mean_feedback_forward_cosine": 0.2933326049854881,
+ "mean_feedback_forward_relative_error": 1.1892362362460087,
+ "min_feedback_forward_cosine": 0.14835697412490845
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 391,
+ "train_examples": 50000,
+ "train_loss": 1.7700924412155152
+ },
+ {
+ "epoch": 2,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.2390552759170532,
+ "mean_feedback_forward_cosine": 0.4218378600321318,
+ "mean_feedback_forward_relative_error": 1.0760179130654586,
+ "min_feedback_forward_cosine": 0.24182388186454773
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 782,
+ "train_examples": 50000,
+ "train_loss": 1.3123860162734986
+ },
+ {
+ "epoch": 3,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.1830275058746338,
+ "mean_feedback_forward_cosine": 0.52299333873548,
+ "mean_feedback_forward_relative_error": 0.9784546917990634,
+ "min_feedback_forward_cosine": 0.3142080307006836
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 1173,
+ "train_examples": 50000,
+ "train_loss": 1.0395409072113038
+ },
+ {
+ "epoch": 4,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.1328173875808716,
+ "mean_feedback_forward_cosine": 0.5936093926429749,
+ "mean_feedback_forward_relative_error": 0.9037119887377086,
+ "min_feedback_forward_cosine": 0.3751407265663147
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 1564,
+ "train_examples": 50000,
+ "train_loss": 0.8746997144889832
+ },
+ {
+ "epoch": 5,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.089064598083496,
+ "mean_feedback_forward_cosine": 0.647214949131012,
+ "mean_feedback_forward_relative_error": 0.8424453154990548,
+ "min_feedback_forward_cosine": 0.42529189586639404
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 1955,
+ "train_examples": 50000,
+ "train_loss": 0.7610496779251099
+ },
+ {
+ "epoch": 6,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.046331524848938,
+ "mean_feedback_forward_cosine": 0.6885352809178201,
+ "mean_feedback_forward_relative_error": 0.7914931131036658,
+ "min_feedback_forward_cosine": 0.47133466601371765
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 2346,
+ "train_examples": 50000,
+ "train_loss": 0.6900432537651062
+ },
+ {
+ "epoch": 7,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 1.006310224533081,
+ "mean_feedback_forward_cosine": 0.7220533270584909,
+ "mean_feedback_forward_relative_error": 0.7474578791543057,
+ "min_feedback_forward_cosine": 0.5120689272880554
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 2737,
+ "train_examples": 50000,
+ "train_loss": 0.638397527961731
+ },
+ {
+ "epoch": 8,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.9711334109306335,
+ "mean_feedback_forward_cosine": 0.7497436592453405,
+ "mean_feedback_forward_relative_error": 0.7091254795852461,
+ "min_feedback_forward_cosine": 0.5461207628250122
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 3128,
+ "train_examples": 50000,
+ "train_loss": 0.5928190508460999
+ },
+ {
+ "epoch": 9,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.9362074732780457,
+ "mean_feedback_forward_cosine": 0.7737372266618829,
+ "mean_feedback_forward_relative_error": 0.6739945019546308,
+ "min_feedback_forward_cosine": 0.578745424747467
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 3519,
+ "train_examples": 50000,
+ "train_loss": 0.5591936241149902
+ },
+ {
+ "epoch": 10,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.9010659456253052,
+ "mean_feedback_forward_cosine": 0.7943896776751468,
+ "mean_feedback_forward_relative_error": 0.6420491651484841,
+ "min_feedback_forward_cosine": 0.6103942394256592
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 3910,
+ "train_examples": 50000,
+ "train_loss": 0.5313378884029388
+ },
+ {
+ "epoch": 11,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.8672091364860535,
+ "mean_feedback_forward_cosine": 0.8132783017660442,
+ "mean_feedback_forward_relative_error": 0.6113202681666926,
+ "min_feedback_forward_cosine": 0.6394283771514893
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 4301,
+ "train_examples": 50000,
+ "train_loss": 0.5099901778316498
+ },
+ {
+ "epoch": 12,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.8353588581085205,
+ "mean_feedback_forward_cosine": 0.8294956621370817,
+ "mean_feedback_forward_relative_error": 0.5838856022608908,
+ "min_feedback_forward_cosine": 0.665457010269165
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 4692,
+ "train_examples": 50000,
+ "train_loss": 0.4878437393760681
+ },
+ {
+ "epoch": 13,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.8048327565193176,
+ "mean_feedback_forward_cosine": 0.8440827444980019,
+ "mean_feedback_forward_relative_error": 0.5580512802851828,
+ "min_feedback_forward_cosine": 0.6900070905685425
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 5083,
+ "train_examples": 50000,
+ "train_loss": 0.46536738251686094
+ },
+ {
+ "epoch": 14,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.7739757299423218,
+ "mean_feedback_forward_cosine": 0.8571237043330544,
+ "mean_feedback_forward_relative_error": 0.5340167738889393,
+ "min_feedback_forward_cosine": 0.7133912444114685
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 5474,
+ "train_examples": 50000,
+ "train_loss": 0.44650153921127317
+ },
+ {
+ "epoch": 15,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.7438143491744995,
+ "mean_feedback_forward_cosine": 0.8689768941778886,
+ "mean_feedback_forward_relative_error": 0.5109366437322215,
+ "min_feedback_forward_cosine": 0.7352336645126343
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 5865,
+ "train_examples": 50000,
+ "train_loss": 0.4332742894744873
+ },
+ {
+ "epoch": 16,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.7148537635803223,
+ "mean_feedback_forward_cosine": 0.8795209809353477,
+ "mean_feedback_forward_relative_error": 0.48969328481900065,
+ "min_feedback_forward_cosine": 0.7554726600646973
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 6256,
+ "train_examples": 50000,
+ "train_loss": 0.4203475615787506
+ },
+ {
+ "epoch": 17,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.6861604452133179,
+ "mean_feedback_forward_cosine": 0.889490290691978,
+ "mean_feedback_forward_relative_error": 0.46875388685025665,
+ "min_feedback_forward_cosine": 0.7745020389556885
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 6647,
+ "train_examples": 50000,
+ "train_loss": 0.40506954431533815
+ },
+ {
+ "epoch": 18,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.6597951650619507,
+ "mean_feedback_forward_cosine": 0.8982065382756685,
+ "mean_feedback_forward_relative_error": 0.4497144394799283,
+ "min_feedback_forward_cosine": 0.7913845777511597
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 7038,
+ "train_examples": 50000,
+ "train_loss": 0.3948731394767761
+ },
+ {
+ "epoch": 19,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.6334631443023682,
+ "mean_feedback_forward_cosine": 0.9063881855261954,
+ "mean_feedback_forward_relative_error": 0.43093959281319066,
+ "min_feedback_forward_cosine": 0.8077483177185059
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 7429,
+ "train_examples": 50000,
+ "train_loss": 0.38731814023971556
+ },
+ {
+ "epoch": 20,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.6084911823272705,
+ "mean_feedback_forward_cosine": 0.9137261102074071,
+ "mean_feedback_forward_relative_error": 0.41337695717811584,
+ "min_feedback_forward_cosine": 0.8225244283676147
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 7820,
+ "train_examples": 50000,
+ "train_loss": 0.3732180533218384
+ },
+ {
+ "epoch": 21,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.5843833684921265,
+ "mean_feedback_forward_cosine": 0.9205317214915627,
+ "mean_feedback_forward_relative_error": 0.3965728541738109,
+ "min_feedback_forward_cosine": 0.8362374305725098
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 8211,
+ "train_examples": 50000,
+ "train_loss": 0.3684745170402527
+ },
+ {
+ "epoch": 22,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.5614224076271057,
+ "mean_feedback_forward_cosine": 0.9267226049774572,
+ "mean_feedback_forward_relative_error": 0.38049717090631785,
+ "min_feedback_forward_cosine": 0.8487249612808228
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 8602,
+ "train_examples": 50000,
+ "train_loss": 0.3560307614326477
+ },
+ {
+ "epoch": 23,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.5392521023750305,
+ "mean_feedback_forward_cosine": 0.9324022406025937,
+ "mean_feedback_forward_relative_error": 0.36515061322011444,
+ "min_feedback_forward_cosine": 0.8603371381759644
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 8993,
+ "train_examples": 50000,
+ "train_loss": 0.34650497359275817
+ },
+ {
+ "epoch": 24,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.5180007219314575,
+ "mean_feedback_forward_cosine": 0.9376050767145658,
+ "mean_feedback_forward_relative_error": 0.3506299837639457,
+ "min_feedback_forward_cosine": 0.8710404634475708
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 9384,
+ "train_examples": 50000,
+ "train_loss": 0.3413775287055969
+ },
+ {
+ "epoch": 25,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.4975259602069855,
+ "mean_feedback_forward_cosine": 0.9422080108993932,
+ "mean_feedback_forward_relative_error": 0.3372772908524463,
+ "min_feedback_forward_cosine": 0.8809491395950317
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 9775,
+ "train_examples": 50000,
+ "train_loss": 0.330313382434845
+ },
+ {
+ "epoch": 26,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.47742199897766113,
+ "mean_feedback_forward_cosine": 0.9466079881316737,
+ "mean_feedback_forward_relative_error": 0.3239913253407729,
+ "min_feedback_forward_cosine": 0.8902804851531982
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 10166,
+ "train_examples": 50000,
+ "train_loss": 0.3268633655643463
+ },
+ {
+ "epoch": 27,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.4579554498195648,
+ "mean_feedback_forward_cosine": 0.9506835184599224,
+ "mean_feedback_forward_relative_error": 0.3112199290802604,
+ "min_feedback_forward_cosine": 0.8989297151565552
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 10557,
+ "train_examples": 50000,
+ "train_loss": 0.32068091130256654
+ },
+ {
+ "epoch": 28,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.4393923580646515,
+ "mean_feedback_forward_cosine": 0.9544056120671724,
+ "mean_feedback_forward_relative_error": 0.2991728476787868,
+ "min_feedback_forward_cosine": 0.906851053237915
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 10948,
+ "train_examples": 50000,
+ "train_loss": 0.31577084115982057
+ },
+ {
+ "epoch": 29,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.4212220013141632,
+ "mean_feedback_forward_cosine": 0.9578661761785808,
+ "mean_feedback_forward_relative_error": 0.28741739415808726,
+ "min_feedback_forward_cosine": 0.9143027067184448
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 11339,
+ "train_examples": 50000,
+ "train_loss": 0.3136876461982727
+ },
+ {
+ "epoch": 30,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.4039841890335083,
+ "mean_feedback_forward_cosine": 0.9609958780439276,
+ "mean_feedback_forward_relative_error": 0.27641204391655166,
+ "min_feedback_forward_cosine": 0.9211227893829346
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 11730,
+ "train_examples": 50000,
+ "train_loss": 0.3026728736782074
+ },
+ {
+ "epoch": 31,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.3873434066772461,
+ "mean_feedback_forward_cosine": 0.9639970754322252,
+ "mean_feedback_forward_relative_error": 0.26545458404641403,
+ "min_feedback_forward_cosine": 0.9273790717124939
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 12121,
+ "train_examples": 50000,
+ "train_loss": 0.30320387191772463
+ },
+ {
+ "epoch": 32,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.37161946296691895,
+ "mean_feedback_forward_cosine": 0.9666794036564074,
+ "mean_feedback_forward_relative_error": 0.2552406650624777,
+ "min_feedback_forward_cosine": 0.9331086277961731
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 12512,
+ "train_examples": 50000,
+ "train_loss": 0.29722497003078463
+ },
+ {
+ "epoch": 33,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.35596978664398193,
+ "mean_feedback_forward_cosine": 0.9692316463119105,
+ "mean_feedback_forward_relative_error": 0.24517358643443962,
+ "min_feedback_forward_cosine": 0.9385503530502319
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 12903,
+ "train_examples": 50000,
+ "train_loss": 0.2951163592720032
+ },
+ {
+ "epoch": 34,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.34164920449256897,
+ "mean_feedback_forward_cosine": 0.9714941225553814,
+ "mean_feedback_forward_relative_error": 0.2358894959876412,
+ "min_feedback_forward_cosine": 0.9433460235595703
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 13294,
+ "train_examples": 50000,
+ "train_loss": 0.2886360764503479
+ },
+ {
+ "epoch": 35,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.3279237151145935,
+ "mean_feedback_forward_cosine": 0.9736838152534083,
+ "mean_feedback_forward_relative_error": 0.22657095170334765,
+ "min_feedback_forward_cosine": 0.9477298259735107
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 13685,
+ "train_examples": 50000,
+ "train_loss": 0.286004674577713
+ },
+ {
+ "epoch": 36,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.3148072361946106,
+ "mean_feedback_forward_cosine": 0.9756333012329904,
+ "mean_feedback_forward_relative_error": 0.21795125815429187,
+ "min_feedback_forward_cosine": 0.9517859220504761
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 14076,
+ "train_examples": 50000,
+ "train_loss": 0.28205583692550656
+ },
+ {
+ "epoch": 37,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.30220264196395874,
+ "mean_feedback_forward_cosine": 0.9774411508911535,
+ "mean_feedback_forward_relative_error": 0.20961935896622508,
+ "min_feedback_forward_cosine": 0.9555267691612244
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 14467,
+ "train_examples": 50000,
+ "train_loss": 0.27561885961532595
+ },
+ {
+ "epoch": 38,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.2898789644241333,
+ "mean_feedback_forward_cosine": 0.9791205744994315,
+ "mean_feedback_forward_relative_error": 0.20157087555057124,
+ "min_feedback_forward_cosine": 0.959039032459259
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 14858,
+ "train_examples": 50000,
+ "train_loss": 0.2754082911109924
+ },
+ {
+ "epoch": 39,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.2784629762172699,
+ "mean_feedback_forward_cosine": 0.9806733633342543,
+ "mean_feedback_forward_relative_error": 0.1938883513212204,
+ "min_feedback_forward_cosine": 0.9621678590774536
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 15249,
+ "train_examples": 50000,
+ "train_loss": 0.27209105748176576
+ },
+ {
+ "epoch": 40,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.2672949731349945,
+ "mean_feedback_forward_cosine": 0.9820656274494372,
+ "mean_feedback_forward_relative_error": 0.18673033188832433,
+ "min_feedback_forward_cosine": 0.9650894999504089
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 15640,
+ "train_examples": 50000,
+ "train_loss": 0.2652701909542084
+ },
+ {
+ "epoch": 41,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.25676229596138,
+ "mean_feedback_forward_cosine": 0.9833770676663047,
+ "mean_feedback_forward_relative_error": 0.17967709624453596,
+ "min_feedback_forward_cosine": 0.9677592515945435
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 16031,
+ "train_examples": 50000,
+ "train_loss": 0.2634839852333069
+ },
+ {
+ "epoch": 42,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.2463054209947586,
+ "mean_feedback_forward_cosine": 0.984612132373609,
+ "mean_feedback_forward_relative_error": 0.17282210486499885,
+ "min_feedback_forward_cosine": 0.9702930450439453
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 16422,
+ "train_examples": 50000,
+ "train_loss": 0.2638785872554779
+ },
+ {
+ "epoch": 43,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.23629648983478546,
+ "mean_feedback_forward_cosine": 0.9857749499772724,
+ "mean_feedback_forward_relative_error": 0.16613267871894336,
+ "min_feedback_forward_cosine": 0.9726332426071167
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 16813,
+ "train_examples": 50000,
+ "train_loss": 0.25648207290649416
+ },
+ {
+ "epoch": 44,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.2265455573797226,
+ "mean_feedback_forward_cosine": 0.9868410668875042,
+ "mean_feedback_forward_relative_error": 0.1597374916860932,
+ "min_feedback_forward_cosine": 0.974839985370636
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 17204,
+ "train_examples": 50000,
+ "train_loss": 0.25590220363616945
+ },
+ {
+ "epoch": 45,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.21779561042785645,
+ "mean_feedback_forward_cosine": 0.9878014420208178,
+ "mean_feedback_forward_relative_error": 0.15373511455560984,
+ "min_feedback_forward_cosine": 0.976719319820404
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 17595,
+ "train_examples": 50000,
+ "train_loss": 0.25498896251678466
+ },
+ {
+ "epoch": 46,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.2088119387626648,
+ "mean_feedback_forward_cosine": 0.9887145381224783,
+ "mean_feedback_forward_relative_error": 0.14787241149889796,
+ "min_feedback_forward_cosine": 0.9785810112953186
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 17986,
+ "train_examples": 50000,
+ "train_loss": 0.25229337076187136
+ },
+ {
+ "epoch": 47,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.20047518610954285,
+ "mean_feedback_forward_cosine": 0.9895525700167606,
+ "mean_feedback_forward_relative_error": 0.14222895373639308,
+ "min_feedback_forward_cosine": 0.980238676071167
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 18377,
+ "train_examples": 50000,
+ "train_loss": 0.25046150511264803
+ },
+ {
+ "epoch": 48,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.19215868413448334,
+ "mean_feedback_forward_cosine": 0.9903604325495268,
+ "mean_feedback_forward_relative_error": 0.13656484335660934,
+ "min_feedback_forward_cosine": 0.9818353652954102
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 18768,
+ "train_examples": 50000,
+ "train_loss": 0.2514414472556114
+ },
+ {
+ "epoch": 49,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.18437601625919342,
+ "mean_feedback_forward_cosine": 0.99109695459667,
+ "mean_feedback_forward_relative_error": 0.13123987673928864,
+ "min_feedback_forward_cosine": 0.9832648634910583
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 19159,
+ "train_examples": 50000,
+ "train_loss": 0.24994389437675477
+ },
+ {
+ "epoch": 50,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.17699532210826874,
+ "mean_feedback_forward_cosine": 0.9917633533477783,
+ "mean_feedback_forward_relative_error": 0.12621399701425903,
+ "min_feedback_forward_cosine": 0.9845625758171082
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 19550,
+ "train_examples": 50000,
+ "train_loss": 0.24948119516849518
+ },
+ {
+ "epoch": 51,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.17000949382781982,
+ "mean_feedback_forward_cosine": 0.9923716909007022,
+ "mean_feedback_forward_relative_error": 0.12141832277963036,
+ "min_feedback_forward_cosine": 0.9857463240623474
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 19941,
+ "train_examples": 50000,
+ "train_loss": 0.2397834491920471
+ },
+ {
+ "epoch": 52,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.16322658956050873,
+ "mean_feedback_forward_cosine": 0.9929383742181879,
+ "mean_feedback_forward_relative_error": 0.11681727555237319,
+ "min_feedback_forward_cosine": 0.9868492484092712
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 20332,
+ "train_examples": 50000,
+ "train_loss": 0.23811577649116517
+ },
+ {
+ "epoch": 53,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.15688201785087585,
+ "mean_feedback_forward_cosine": 0.9934662298152321,
+ "mean_feedback_forward_relative_error": 0.11234608331793233,
+ "min_feedback_forward_cosine": 0.9878503084182739
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 20723,
+ "train_examples": 50000,
+ "train_loss": 0.2347814343738556
+ },
+ {
+ "epoch": 54,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.1504819542169571,
+ "mean_feedback_forward_cosine": 0.9939507340130053,
+ "mean_feedback_forward_relative_error": 0.10809194061316942,
+ "min_feedback_forward_cosine": 0.9888131022453308
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 21114,
+ "train_examples": 50000,
+ "train_loss": 0.24166065438747406
+ },
+ {
+ "epoch": 55,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.14444631338119507,
+ "mean_feedback_forward_cosine": 0.9944124911960802,
+ "mean_feedback_forward_relative_error": 0.10387111808124341,
+ "min_feedback_forward_cosine": 0.9896870851516724
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 21505,
+ "train_examples": 50000,
+ "train_loss": 0.2370883663749695
+ },
+ {
+ "epoch": 56,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.13871312141418457,
+ "mean_feedback_forward_cosine": 0.9948335038988214,
+ "mean_feedback_forward_relative_error": 0.0998734708287214,
+ "min_feedback_forward_cosine": 0.990486741065979
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 21896,
+ "train_examples": 50000,
+ "train_loss": 0.23285125886917113
+ },
+ {
+ "epoch": 57,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.1331096887588501,
+ "mean_feedback_forward_cosine": 0.9952221042231509,
+ "mean_feedback_forward_relative_error": 0.09602151222919163,
+ "min_feedback_forward_cosine": 0.9912351369857788
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 22287,
+ "train_examples": 50000,
+ "train_loss": 0.23394198512077333
+ },
+ {
+ "epoch": 58,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.12772555649280548,
+ "mean_feedback_forward_cosine": 0.9955923525910628,
+ "mean_feedback_forward_relative_error": 0.09223919832392742,
+ "min_feedback_forward_cosine": 0.9919288158416748
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 22678,
+ "train_examples": 50000,
+ "train_loss": 0.23523316113471984
+ },
+ {
+ "epoch": 59,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.12264999002218246,
+ "mean_feedback_forward_cosine": 0.9959284349491722,
+ "mean_feedback_forward_relative_error": 0.08864477080734152,
+ "min_feedback_forward_cosine": 0.9925565123558044
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 23069,
+ "train_examples": 50000,
+ "train_loss": 0.22811366169929503
+ },
+ {
+ "epoch": 60,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.1177108958363533,
+ "mean_feedback_forward_cosine": 0.9962369297680101,
+ "mean_feedback_forward_relative_error": 0.0852269875375848,
+ "min_feedback_forward_cosine": 0.9931412935256958
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 23460,
+ "train_examples": 50000,
+ "train_loss": 0.227693050532341
+ },
+ {
+ "epoch": 61,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.11301573365926743,
+ "mean_feedback_forward_cosine": 0.9965199357584903,
+ "mean_feedback_forward_relative_error": 0.08195626108269942,
+ "min_feedback_forward_cosine": 0.9936733841896057
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 23851,
+ "train_examples": 50000,
+ "train_loss": 0.22571406733989716
+ },
+ {
+ "epoch": 62,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.10847260802984238,
+ "mean_feedback_forward_cosine": 0.9967795798653051,
+ "mean_feedback_forward_relative_error": 0.07883979870300543,
+ "min_feedback_forward_cosine": 0.9941681623458862
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 24242,
+ "train_examples": 50000,
+ "train_loss": 0.22897239991188048
+ },
+ {
+ "epoch": 63,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.10419360548257828,
+ "mean_feedback_forward_cosine": 0.9970179796218872,
+ "mean_feedback_forward_relative_error": 0.07585975821865232,
+ "min_feedback_forward_cosine": 0.9946166276931763
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 24633,
+ "train_examples": 50000,
+ "train_loss": 0.22848598481178284
+ },
+ {
+ "epoch": 64,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.1000451073050499,
+ "mean_feedback_forward_cosine": 0.9972450827297411,
+ "mean_feedback_forward_relative_error": 0.07290914213578951,
+ "min_feedback_forward_cosine": 0.9950348138809204
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 25024,
+ "train_examples": 50000,
+ "train_loss": 0.22550214426994325
+ },
+ {
+ "epoch": 65,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.09609720855951309,
+ "mean_feedback_forward_cosine": 0.9974487361155058,
+ "mean_feedback_forward_relative_error": 0.07015403369931798,
+ "min_feedback_forward_cosine": 0.9954178333282471
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 25415,
+ "train_examples": 50000,
+ "train_loss": 0.22258466155052184
+ },
+ {
+ "epoch": 66,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.09232494235038757,
+ "mean_feedback_forward_cosine": 0.9976408418856169,
+ "mean_feedback_forward_relative_error": 0.06746580520350683,
+ "min_feedback_forward_cosine": 0.9957685470581055
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 25806,
+ "train_examples": 50000,
+ "train_loss": 0.21734946195602417
+ },
+ {
+ "epoch": 67,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.08847074955701828,
+ "mean_feedback_forward_cosine": 0.9978232603324088,
+ "mean_feedback_forward_relative_error": 0.0648055956944039,
+ "min_feedback_forward_cosine": 0.9961132407188416
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 26197,
+ "train_examples": 50000,
+ "train_loss": 0.2291919409275055
+ },
+ {
+ "epoch": 68,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.08493213355541229,
+ "mean_feedback_forward_cosine": 0.997984418743535,
+ "mean_feedback_forward_relative_error": 0.062354145963725294,
+ "min_feedback_forward_cosine": 0.9964166879653931
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 26588,
+ "train_examples": 50000,
+ "train_loss": 0.21789681928634644
+ },
+ {
+ "epoch": 69,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.08157175034284592,
+ "mean_feedback_forward_cosine": 0.9981369501666019,
+ "mean_feedback_forward_relative_error": 0.05995147479207892,
+ "min_feedback_forward_cosine": 0.9966939687728882
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 26979,
+ "train_examples": 50000,
+ "train_loss": 0.2237079916858673
+ },
+ {
+ "epoch": 70,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0784633606672287,
+ "mean_feedback_forward_cosine": 0.9982732910858957,
+ "mean_feedback_forward_relative_error": 0.05771061552590445,
+ "min_feedback_forward_cosine": 0.9969395399093628
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 27370,
+ "train_examples": 50000,
+ "train_loss": 0.21691304666519165
+ },
+ {
+ "epoch": 71,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.07528278976678848,
+ "mean_feedback_forward_cosine": 0.9984068431352314,
+ "mean_feedback_forward_relative_error": 0.05543650039716771,
+ "min_feedback_forward_cosine": 0.9971820116043091
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 27761,
+ "train_examples": 50000,
+ "train_loss": 0.2197783943271637
+ },
+ {
+ "epoch": 72,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.07251707464456558,
+ "mean_feedback_forward_cosine": 0.9985218236320897,
+ "mean_feedback_forward_relative_error": 0.05338385563931967,
+ "min_feedback_forward_cosine": 0.9973851442337036
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 28152,
+ "train_examples": 50000,
+ "train_loss": 0.20860771854400634
+ },
+ {
+ "epoch": 73,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.06950698047876358,
+ "mean_feedback_forward_cosine": 0.9986368041289481,
+ "mean_feedback_forward_relative_error": 0.05126833121635412,
+ "min_feedback_forward_cosine": 0.997597336769104
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 28543,
+ "train_examples": 50000,
+ "train_loss": 0.22113546739578246
+ },
+ {
+ "epoch": 74,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.06674308329820633,
+ "mean_feedback_forward_cosine": 0.9987422478826422,
+ "mean_feedback_forward_relative_error": 0.04924817549947061,
+ "min_feedback_forward_cosine": 0.9977837800979614
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 28934,
+ "train_examples": 50000,
+ "train_loss": 0.21520923093795777
+ },
+ {
+ "epoch": 75,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.06415430456399918,
+ "mean_feedback_forward_cosine": 0.9988358365861993,
+ "mean_feedback_forward_relative_error": 0.0473746385817465,
+ "min_feedback_forward_cosine": 0.9979520440101624
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 29325,
+ "train_examples": 50000,
+ "train_loss": 0.21333048491954804
+ },
+ {
+ "epoch": 76,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.06168223172426224,
+ "mean_feedback_forward_cosine": 0.9989237346147236,
+ "mean_feedback_forward_relative_error": 0.045554709571756814,
+ "min_feedback_forward_cosine": 0.9981064796447754
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 29716,
+ "train_examples": 50000,
+ "train_loss": 0.21425045120239258
+ },
+ {
+ "epoch": 77,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.05932221561670303,
+ "mean_feedback_forward_cosine": 0.9990047937945316,
+ "mean_feedback_forward_relative_error": 0.04380172589107564,
+ "min_feedback_forward_cosine": 0.9982483386993408
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 30107,
+ "train_examples": 50000,
+ "train_loss": 0.20893779522418976
+ },
+ {
+ "epoch": 78,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.057064302265644073,
+ "mean_feedback_forward_cosine": 0.9990780792738262,
+ "mean_feedback_forward_relative_error": 0.042154407618861446,
+ "min_feedback_forward_cosine": 0.9983787536621094
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 30498,
+ "train_examples": 50000,
+ "train_loss": 0.20841411416530609
+ },
+ {
+ "epoch": 79,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.05485239252448082,
+ "mean_feedback_forward_cosine": 0.9991467187279149,
+ "mean_feedback_forward_relative_error": 0.04055123815411016,
+ "min_feedback_forward_cosine": 0.9985014200210571
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 30889,
+ "train_examples": 50000,
+ "train_loss": 0.20558881327152254
+ },
+ {
+ "epoch": 80,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.05262824147939682,
+ "mean_feedback_forward_cosine": 0.9992108250919142,
+ "mean_feedback_forward_relative_error": 0.03899813884575116,
+ "min_feedback_forward_cosine": 0.9986202120780945
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 31280,
+ "train_examples": 50000,
+ "train_loss": 0.20863534938812256
+ },
+ {
+ "epoch": 81,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.05054010823369026,
+ "mean_feedback_forward_cosine": 0.9992698587869343,
+ "mean_feedback_forward_relative_error": 0.03751169733311001,
+ "min_feedback_forward_cosine": 0.9987270832061768
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 31671,
+ "train_examples": 50000,
+ "train_loss": 0.20875182423114777
+ },
+ {
+ "epoch": 82,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.04847333952784538,
+ "mean_feedback_forward_cosine": 0.9993273270757574,
+ "mean_feedback_forward_relative_error": 0.036009292363336214,
+ "min_feedback_forward_cosine": 0.9988292455673218
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 32062,
+ "train_examples": 50000,
+ "train_loss": 0.21284654743671416
+ },
+ {
+ "epoch": 83,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.04656003415584564,
+ "mean_feedback_forward_cosine": 0.9993772318488673,
+ "mean_feedback_forward_relative_error": 0.03464415968444787,
+ "min_feedback_forward_cosine": 0.9989197850227356
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 32453,
+ "train_examples": 50000,
+ "train_loss": 0.20589884860992433
+ },
+ {
+ "epoch": 84,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.044718120247125626,
+ "mean_feedback_forward_cosine": 0.9994242944215473,
+ "mean_feedback_forward_relative_error": 0.033306461856945566,
+ "min_feedback_forward_cosine": 0.9990034103393555
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 32844,
+ "train_examples": 50000,
+ "train_loss": 0.2039555728626251
+ },
+ {
+ "epoch": 85,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.04293333739042282,
+ "mean_feedback_forward_cosine": 0.9994691735819766,
+ "mean_feedback_forward_relative_error": 0.03198307463409085,
+ "min_feedback_forward_cosine": 0.9990812540054321
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 33235,
+ "train_examples": 50000,
+ "train_loss": 0.2055033529663086
+ },
+ {
+ "epoch": 86,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.041266269981861115,
+ "mean_feedback_forward_cosine": 0.9995089016462627,
+ "mean_feedback_forward_relative_error": 0.030763930886199598,
+ "min_feedback_forward_cosine": 0.9991511702537537
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 33626,
+ "train_examples": 50000,
+ "train_loss": 0.20926678663253784
+ },
+ {
+ "epoch": 87,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03962496295571327,
+ "mean_feedback_forward_cosine": 0.9995468196115995,
+ "mean_feedback_forward_relative_error": 0.02955206253222729,
+ "min_feedback_forward_cosine": 0.9992172122001648
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 34017,
+ "train_examples": 50000,
+ "train_loss": 0.20789328810691834
+ },
+ {
+ "epoch": 88,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03799188882112503,
+ "mean_feedback_forward_cosine": 0.9995815001035991,
+ "mean_feedback_forward_relative_error": 0.02840322393335794,
+ "min_feedback_forward_cosine": 0.9992803931236267
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 34408,
+ "train_examples": 50000,
+ "train_loss": 0.21184086050510406
+ },
+ {
+ "epoch": 89,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03658031299710274,
+ "mean_feedback_forward_cosine": 0.9996119706254256,
+ "mean_feedback_forward_relative_error": 0.027347941441755546,
+ "min_feedback_forward_cosine": 0.999332845211029
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 34799,
+ "train_examples": 50000,
+ "train_loss": 0.20259064952373504
+ },
+ {
+ "epoch": 90,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03519190847873688,
+ "mean_feedback_forward_cosine": 0.9996404459601954,
+ "mean_feedback_forward_relative_error": 0.026322352945020323,
+ "min_feedback_forward_cosine": 0.9993824362754822
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 35190,
+ "train_examples": 50000,
+ "train_loss": 0.19757543837547303
+ },
+ {
+ "epoch": 91,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03381280228495598,
+ "mean_feedback_forward_cosine": 0.9996674562755384,
+ "mean_feedback_forward_relative_error": 0.025314428206337124,
+ "min_feedback_forward_cosine": 0.9994298219680786
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 35581,
+ "train_examples": 50000,
+ "train_loss": 0.20042255244255067
+ },
+ {
+ "epoch": 92,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03247952088713646,
+ "mean_feedback_forward_cosine": 0.9996924023879202,
+ "mean_feedback_forward_relative_error": 0.024345717539912777,
+ "min_feedback_forward_cosine": 0.9994738698005676
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 35972,
+ "train_examples": 50000,
+ "train_loss": 0.20054024726867675
+ },
+ {
+ "epoch": 93,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.03116488642990589,
+ "mean_feedback_forward_cosine": 0.9997159681822124,
+ "mean_feedback_forward_relative_error": 0.023396770371810385,
+ "min_feedback_forward_cosine": 0.9995157122612
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 36363,
+ "train_examples": 50000,
+ "train_loss": 0.20490990519046784
+ },
+ {
+ "epoch": 94,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02997639775276184,
+ "mean_feedback_forward_cosine": 0.9997366541310361,
+ "mean_feedback_forward_relative_error": 0.022526048957125136,
+ "min_feedback_forward_cosine": 0.9995517730712891
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 36754,
+ "train_examples": 50000,
+ "train_loss": 0.19781552631139757
+ },
+ {
+ "epoch": 95,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02878512814640999,
+ "mean_feedback_forward_cosine": 0.9997561793578299,
+ "mean_feedback_forward_relative_error": 0.021671489910467676,
+ "min_feedback_forward_cosine": 0.9995867013931274
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 37145,
+ "train_examples": 50000,
+ "train_loss": 0.20112258635997773
+ },
+ {
+ "epoch": 96,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02762950211763382,
+ "mean_feedback_forward_cosine": 0.9997752214732923,
+ "mean_feedback_forward_relative_error": 0.020811174751112337,
+ "min_feedback_forward_cosine": 0.9996191263198853
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 37536,
+ "train_examples": 50000,
+ "train_loss": 0.19829977237701416
+ },
+ {
+ "epoch": 97,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.026539549231529236,
+ "mean_feedback_forward_cosine": 0.9997924252560264,
+ "mean_feedback_forward_relative_error": 0.020000331888073368,
+ "min_feedback_forward_cosine": 0.9996485710144043
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 37927,
+ "train_examples": 50000,
+ "train_loss": 0.20181411871910096
+ },
+ {
+ "epoch": 98,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.025494523346424103,
+ "mean_feedback_forward_cosine": 0.9998080573583904,
+ "mean_feedback_forward_relative_error": 0.0192308844753394,
+ "min_feedback_forward_cosine": 0.9996756911277771
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 38318,
+ "train_examples": 50000,
+ "train_loss": 0.19922858103752136
+ },
+ {
+ "epoch": 99,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02450042963027954,
+ "mean_feedback_forward_cosine": 0.9998224973678589,
+ "mean_feedback_forward_relative_error": 0.018493761112423318,
+ "min_feedback_forward_cosine": 0.9997004866600037
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 38709,
+ "train_examples": 50000,
+ "train_loss": 0.19956167534828187
+ },
+ {
+ "epoch": 100,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02355109713971615,
+ "mean_feedback_forward_cosine": 0.9998359052758468,
+ "mean_feedback_forward_relative_error": 0.01778221743083314,
+ "min_feedback_forward_cosine": 0.9997232556343079
+ },
+ "lr": 0.1,
+ "output_lr": 0.1,
+ "step": 39100,
+ "train_examples": 50000,
+ "train_loss": 0.19581967149734497
+ },
+ {
+ "epoch": 101,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023542871698737144,
+ "mean_feedback_forward_cosine": 0.9998360401705692,
+ "mean_feedback_forward_relative_error": 0.017772498883699115,
+ "min_feedback_forward_cosine": 0.9997234344482422
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 39491,
+ "train_examples": 50000,
+ "train_loss": 0.1236756948375702
+ },
+ {
+ "epoch": 102,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02353820390999317,
+ "mean_feedback_forward_cosine": 0.9998361154606468,
+ "mean_feedback_forward_relative_error": 0.01776506594921413,
+ "min_feedback_forward_cosine": 0.9997235536575317
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 39882,
+ "train_examples": 50000,
+ "train_loss": 0.09747550456523896
+ },
+ {
+ "epoch": 103,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023533513769507408,
+ "mean_feedback_forward_cosine": 0.9998362221215901,
+ "mean_feedback_forward_relative_error": 0.017757885746265713,
+ "min_feedback_forward_cosine": 0.9997236728668213
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 40273,
+ "train_examples": 50000,
+ "train_loss": 0.08808807847499847
+ },
+ {
+ "epoch": 104,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023529035970568657,
+ "mean_feedback_forward_cosine": 0.9998363099600139,
+ "mean_feedback_forward_relative_error": 0.01775112586390031,
+ "min_feedback_forward_cosine": 0.9997237920761108
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 40664,
+ "train_examples": 50000,
+ "train_loss": 0.0813669945693016
+ },
+ {
+ "epoch": 105,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.0235243309289217,
+ "mean_feedback_forward_cosine": 0.9998363789759184,
+ "mean_feedback_forward_relative_error": 0.017744342453385655,
+ "min_feedback_forward_cosine": 0.9997239112854004
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 41055,
+ "train_examples": 50000,
+ "train_loss": 0.07782446749687195
+ },
+ {
+ "epoch": 106,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023519985377788544,
+ "mean_feedback_forward_cosine": 0.9998364856368617,
+ "mean_feedback_forward_relative_error": 0.01773788101088844,
+ "min_feedback_forward_cosine": 0.9997239112854004
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 41446,
+ "train_examples": 50000,
+ "train_loss": 0.07170286382913589
+ },
+ {
+ "epoch": 107,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02351556345820427,
+ "mean_feedback_forward_cosine": 0.9998365515156796,
+ "mean_feedback_forward_relative_error": 0.01773151892580484,
+ "min_feedback_forward_cosine": 0.9997240304946899
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 41837,
+ "train_examples": 50000,
+ "train_loss": 0.06847817730426789
+ },
+ {
+ "epoch": 108,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023511026054620743,
+ "mean_feedback_forward_cosine": 0.9998366330799303,
+ "mean_feedback_forward_relative_error": 0.017725305901350158,
+ "min_feedback_forward_cosine": 0.999724268913269
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 42228,
+ "train_examples": 50000,
+ "train_loss": 0.06606978836536408
+ },
+ {
+ "epoch": 109,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023506496101617813,
+ "mean_feedback_forward_cosine": 0.9998367083700079,
+ "mean_feedback_forward_relative_error": 0.017719032561504526,
+ "min_feedback_forward_cosine": 0.999724268913269
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 42619,
+ "train_examples": 50000,
+ "train_loss": 0.06365371140718461
+ },
+ {
+ "epoch": 110,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023502036929130554,
+ "mean_feedback_forward_cosine": 0.9998367742488259,
+ "mean_feedback_forward_relative_error": 0.01771265836922746,
+ "min_feedback_forward_cosine": 0.9997243881225586
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 43010,
+ "train_examples": 50000,
+ "train_loss": 0.06021396337866783
+ },
+ {
+ "epoch": 111,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02349746786057949,
+ "mean_feedback_forward_cosine": 0.9998368777726826,
+ "mean_feedback_forward_relative_error": 0.01770649887131233,
+ "min_feedback_forward_cosine": 0.9997245073318481
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 43401,
+ "train_examples": 50000,
+ "train_loss": 0.05857227984309196
+ },
+ {
+ "epoch": 112,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02349289506673813,
+ "mean_feedback_forward_cosine": 0.9998369844336259,
+ "mean_feedback_forward_relative_error": 0.01770027577386875,
+ "min_feedback_forward_cosine": 0.9997247457504272
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 43792,
+ "train_examples": 50000,
+ "train_loss": 0.05728690538823605
+ },
+ {
+ "epoch": 113,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023488439619541168,
+ "mean_feedback_forward_cosine": 0.9998370565866169,
+ "mean_feedback_forward_relative_error": 0.01769417431205511,
+ "min_feedback_forward_cosine": 0.9997247457504272
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 44183,
+ "train_examples": 50000,
+ "train_loss": 0.054087092715501786
+ },
+ {
+ "epoch": 114,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023483794182538986,
+ "mean_feedback_forward_cosine": 0.9998371632475602,
+ "mean_feedback_forward_relative_error": 0.017688091672760874,
+ "min_feedback_forward_cosine": 0.9997248649597168
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 44574,
+ "train_examples": 50000,
+ "train_loss": 0.05327781386137009
+ },
+ {
+ "epoch": 115,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023479023948311806,
+ "mean_feedback_forward_cosine": 0.9998372291263781,
+ "mean_feedback_forward_relative_error": 0.017681793236222706,
+ "min_feedback_forward_cosine": 0.9997249841690063
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 44965,
+ "train_examples": 50000,
+ "train_loss": 0.05246301278233528
+ },
+ {
+ "epoch": 116,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02347436733543873,
+ "mean_feedback_forward_cosine": 0.9998373138277155,
+ "mean_feedback_forward_relative_error": 0.0176756494482489,
+ "min_feedback_forward_cosine": 0.9997251033782959
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 45356,
+ "train_examples": 50000,
+ "train_loss": 0.050172818863987924
+ },
+ {
+ "epoch": 117,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02346990443766117,
+ "mean_feedback_forward_cosine": 0.9998373985290527,
+ "mean_feedback_forward_relative_error": 0.017669501101696176,
+ "min_feedback_forward_cosine": 0.9997251629829407
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 45747,
+ "train_examples": 50000,
+ "train_loss": 0.04647394307613373
+ },
+ {
+ "epoch": 118,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023465104401111603,
+ "mean_feedback_forward_cosine": 0.9998374957787363,
+ "mean_feedback_forward_relative_error": 0.017663119850974334,
+ "min_feedback_forward_cosine": 0.9997252821922302
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 46138,
+ "train_examples": 50000,
+ "train_loss": 0.0468657146859169
+ },
+ {
+ "epoch": 119,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02346024103462696,
+ "mean_feedback_forward_cosine": 0.9998375898913333,
+ "mean_feedback_forward_relative_error": 0.01765680999348038,
+ "min_feedback_forward_cosine": 0.9997254610061646
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 46529,
+ "train_examples": 50000,
+ "train_loss": 0.04547766117334366
+ },
+ {
+ "epoch": 120,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023455336689949036,
+ "mean_feedback_forward_cosine": 0.9998376902781034,
+ "mean_feedback_forward_relative_error": 0.01765055241259305,
+ "min_feedback_forward_cosine": 0.9997255802154541
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 46920,
+ "train_examples": 50000,
+ "train_loss": 0.04480087527751923
+ },
+ {
+ "epoch": 121,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023450380191206932,
+ "mean_feedback_forward_cosine": 0.9998377875277871,
+ "mean_feedback_forward_relative_error": 0.017644206919756374,
+ "min_feedback_forward_cosine": 0.9997256398200989
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 47311,
+ "train_examples": 50000,
+ "train_loss": 0.044035287322402
+ },
+ {
+ "epoch": 122,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023445529863238335,
+ "mean_feedback_forward_cosine": 0.9998378659549513,
+ "mean_feedback_forward_relative_error": 0.01763799686082884,
+ "min_feedback_forward_cosine": 0.9997258186340332
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 47702,
+ "train_examples": 50000,
+ "train_loss": 0.0424709809422493
+ },
+ {
+ "epoch": 123,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023440571501851082,
+ "mean_feedback_forward_cosine": 0.9998379443821154,
+ "mean_feedback_forward_relative_error": 0.01763170847277108,
+ "min_feedback_forward_cosine": 0.9997259378433228
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 48093,
+ "train_examples": 50000,
+ "train_loss": 0.04040559020280838
+ },
+ {
+ "epoch": 124,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02343549020588398,
+ "mean_feedback_forward_cosine": 0.9998380604543184,
+ "mean_feedback_forward_relative_error": 0.017625319624417705,
+ "min_feedback_forward_cosine": 0.9997260570526123
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 48484,
+ "train_examples": 50000,
+ "train_loss": 0.0407452875483036
+ },
+ {
+ "epoch": 125,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023430390283465385,
+ "mean_feedback_forward_cosine": 0.9998381545669154,
+ "mean_feedback_forward_relative_error": 0.017618816885116854,
+ "min_feedback_forward_cosine": 0.9997261762619019
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 48875,
+ "train_examples": 50000,
+ "train_loss": 0.0408136903834343
+ },
+ {
+ "epoch": 126,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023425184190273285,
+ "mean_feedback_forward_cosine": 0.9998382549536856,
+ "mean_feedback_forward_relative_error": 0.017612327037281113,
+ "min_feedback_forward_cosine": 0.9997262358665466
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 49266,
+ "train_examples": 50000,
+ "train_loss": 0.03979753321647644
+ },
+ {
+ "epoch": 127,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023419927805662155,
+ "mean_feedback_forward_cosine": 0.9998383427921095,
+ "mean_feedback_forward_relative_error": 0.017605616760097052,
+ "min_feedback_forward_cosine": 0.9997263550758362
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 49657,
+ "train_examples": 50000,
+ "train_loss": 0.03905742102980614
+ },
+ {
+ "epoch": 128,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023414863273501396,
+ "mean_feedback_forward_cosine": 0.9998384525901393,
+ "mean_feedback_forward_relative_error": 0.017599164238689763,
+ "min_feedback_forward_cosine": 0.9997264742851257
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 50048,
+ "train_examples": 50000,
+ "train_loss": 0.036579988394975664
+ },
+ {
+ "epoch": 129,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023409733548760414,
+ "mean_feedback_forward_cosine": 0.9998385717994288,
+ "mean_feedback_forward_relative_error": 0.0175927120849098,
+ "min_feedback_forward_cosine": 0.9997266530990601
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 50439,
+ "train_examples": 50000,
+ "train_loss": 0.03675649410486221
+ },
+ {
+ "epoch": 130,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02340470626950264,
+ "mean_feedback_forward_cosine": 0.9998386533636796,
+ "mean_feedback_forward_relative_error": 0.017586188292817065,
+ "min_feedback_forward_cosine": 0.9997266530990601
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 50830,
+ "train_examples": 50000,
+ "train_loss": 0.03424716347932816
+ },
+ {
+ "epoch": 131,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02339932695031166,
+ "mean_feedback_forward_cosine": 0.9998387788471422,
+ "mean_feedback_forward_relative_error": 0.017579433924861644,
+ "min_feedback_forward_cosine": 0.9997269511222839
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 51221,
+ "train_examples": 50000,
+ "train_loss": 0.03485953302919865
+ },
+ {
+ "epoch": 132,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02339424192905426,
+ "mean_feedback_forward_cosine": 0.9998388729597393,
+ "mean_feedback_forward_relative_error": 0.017572860699146986,
+ "min_feedback_forward_cosine": 0.9997268915176392
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 51612,
+ "train_examples": 50000,
+ "train_loss": 0.03307149721741676
+ },
+ {
+ "epoch": 133,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02338869497179985,
+ "mean_feedback_forward_cosine": 0.9998389576610766,
+ "mean_feedback_forward_relative_error": 0.017566105988072723,
+ "min_feedback_forward_cosine": 0.9997270703315735
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 52003,
+ "train_examples": 50000,
+ "train_loss": 0.03417502753615379
+ },
+ {
+ "epoch": 134,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02338322438299656,
+ "mean_feedback_forward_cosine": 0.9998390800074527,
+ "mean_feedback_forward_relative_error": 0.017559204740743888,
+ "min_feedback_forward_cosine": 0.9997271299362183
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 52394,
+ "train_examples": 50000,
+ "train_loss": 0.03339349335074425
+ },
+ {
+ "epoch": 135,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023377852514386177,
+ "mean_feedback_forward_cosine": 0.9998391709829632,
+ "mean_feedback_forward_relative_error": 0.017552312953691734,
+ "min_feedback_forward_cosine": 0.9997272491455078
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 52785,
+ "train_examples": 50000,
+ "train_loss": 0.03251529821872711
+ },
+ {
+ "epoch": 136,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023372720927000046,
+ "mean_feedback_forward_cosine": 0.999839302740599,
+ "mean_feedback_forward_relative_error": 0.01754566536922204,
+ "min_feedback_forward_cosine": 0.9997274875640869
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 53176,
+ "train_examples": 50000,
+ "train_loss": 0.030737761940956117
+ },
+ {
+ "epoch": 137,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02336716465651989,
+ "mean_feedback_forward_cosine": 0.9998394062644557,
+ "mean_feedback_forward_relative_error": 0.01753867660207968,
+ "min_feedback_forward_cosine": 0.9997276067733765
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 53567,
+ "train_examples": 50000,
+ "train_loss": 0.031349986025094985
+ },
+ {
+ "epoch": 138,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023361675441265106,
+ "mean_feedback_forward_cosine": 0.9998395286108318,
+ "mean_feedback_forward_relative_error": 0.01753187581504646,
+ "min_feedback_forward_cosine": 0.999727725982666
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 53958,
+ "train_examples": 50000,
+ "train_loss": 0.029603690266013144
+ },
+ {
+ "epoch": 139,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02335577830672264,
+ "mean_feedback_forward_cosine": 0.9998395976267362,
+ "mean_feedback_forward_relative_error": 0.017524560888935076,
+ "min_feedback_forward_cosine": 0.9997277855873108
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 54349,
+ "train_examples": 50000,
+ "train_loss": 0.03143566385626793
+ },
+ {
+ "epoch": 140,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023349737748503685,
+ "mean_feedback_forward_cosine": 0.9998397733035841,
+ "mean_feedback_forward_relative_error": 0.01751705925715597,
+ "min_feedback_forward_cosine": 0.9997280240058899
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 54740,
+ "train_examples": 50000,
+ "train_loss": 0.031357577148675916
+ },
+ {
+ "epoch": 141,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023343779146671295,
+ "mean_feedback_forward_cosine": 0.9998398768274408,
+ "mean_feedback_forward_relative_error": 0.01750972472425354,
+ "min_feedback_forward_cosine": 0.9997281432151794
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 55131,
+ "train_examples": 50000,
+ "train_loss": 0.030024923699498177
+ },
+ {
+ "epoch": 142,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023337777704000473,
+ "mean_feedback_forward_cosine": 0.9998399866254706,
+ "mean_feedback_forward_relative_error": 0.01750226997721352,
+ "min_feedback_forward_cosine": 0.999728262424469
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 55522,
+ "train_examples": 50000,
+ "train_loss": 0.02923239346385002
+ },
+ {
+ "epoch": 143,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023331688717007637,
+ "mean_feedback_forward_cosine": 0.9998400964235005,
+ "mean_feedback_forward_relative_error": 0.017494787805174526,
+ "min_feedback_forward_cosine": 0.9997283816337585
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 55913,
+ "train_examples": 50000,
+ "train_loss": 0.028809776036143302
+ },
+ {
+ "epoch": 144,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023325927555561066,
+ "mean_feedback_forward_cosine": 0.9998402093586168,
+ "mean_feedback_forward_relative_error": 0.017487642526822656,
+ "min_feedback_forward_cosine": 0.9997285604476929
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 56304,
+ "train_examples": 50000,
+ "train_loss": 0.027221469062566755
+ },
+ {
+ "epoch": 145,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023319661617279053,
+ "mean_feedback_forward_cosine": 0.9998403473904258,
+ "mean_feedback_forward_relative_error": 0.01748000379455717,
+ "min_feedback_forward_cosine": 0.9997286200523376
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 56695,
+ "train_examples": 50000,
+ "train_loss": 0.028306029340326787
+ },
+ {
+ "epoch": 146,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023313837125897408,
+ "mean_feedback_forward_cosine": 0.9998404634626288,
+ "mean_feedback_forward_relative_error": 0.01747281150892377,
+ "min_feedback_forward_cosine": 0.999728798866272
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 57086,
+ "train_examples": 50000,
+ "train_loss": 0.02637038481593132
+ },
+ {
+ "epoch": 147,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023307349532842636,
+ "mean_feedback_forward_cosine": 0.9998405983573512,
+ "mean_feedback_forward_relative_error": 0.017465155326614256,
+ "min_feedback_forward_cosine": 0.9997289776802063
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 57477,
+ "train_examples": 50000,
+ "train_loss": 0.02744144564986229
+ },
+ {
+ "epoch": 148,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023300573229789734,
+ "mean_feedback_forward_cosine": 0.9998407144295541,
+ "mean_feedback_forward_relative_error": 0.017457119140185808,
+ "min_feedback_forward_cosine": 0.9997292160987854
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 57868,
+ "train_examples": 50000,
+ "train_loss": 0.027158371942043305
+ },
+ {
+ "epoch": 149,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023294003680348396,
+ "mean_feedback_forward_cosine": 0.9998408273646706,
+ "mean_feedback_forward_relative_error": 0.017449136823415756,
+ "min_feedback_forward_cosine": 0.9997292757034302
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 58259,
+ "train_examples": 50000,
+ "train_loss": 0.027715308804512024
+ },
+ {
+ "epoch": 150,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023287752643227577,
+ "mean_feedback_forward_cosine": 0.9998409779448258,
+ "mean_feedback_forward_relative_error": 0.017441295283405406,
+ "min_feedback_forward_cosine": 0.9997295141220093
+ },
+ "lr": 0.010000000000000002,
+ "output_lr": 0.010000000000000002,
+ "step": 58650,
+ "train_examples": 50000,
+ "train_loss": 0.025534065507054327
+ },
+ {
+ "epoch": 151,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023287681862711906,
+ "mean_feedback_forward_cosine": 0.9998409716706527,
+ "mean_feedback_forward_relative_error": 0.01744110637197369,
+ "min_feedback_forward_cosine": 0.9997294545173645
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 59041,
+ "train_examples": 50000,
+ "train_loss": 0.023126552869677543
+ },
+ {
+ "epoch": 152,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023287629708647728,
+ "mean_feedback_forward_cosine": 0.9998409748077393,
+ "mean_feedback_forward_relative_error": 0.017440940204419587,
+ "min_feedback_forward_cosine": 0.9997293949127197
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 59432,
+ "train_examples": 50000,
+ "train_loss": 0.02281324056327343
+ },
+ {
+ "epoch": 153,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023287585005164146,
+ "mean_feedback_forward_cosine": 0.9998409716706527,
+ "mean_feedback_forward_relative_error": 0.01744077724747752,
+ "min_feedback_forward_cosine": 0.9997295141220093
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 59823,
+ "train_examples": 50000,
+ "train_loss": 0.021053250761926174
+ },
+ {
+ "epoch": 154,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023287536576390266,
+ "mean_feedback_forward_cosine": 0.9998409842189989,
+ "mean_feedback_forward_relative_error": 0.017440615172841047,
+ "min_feedback_forward_cosine": 0.9997294545173645
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 60214,
+ "train_examples": 50000,
+ "train_loss": 0.021394379252716898
+ },
+ {
+ "epoch": 155,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023287488147616386,
+ "mean_feedback_forward_cosine": 0.9998409999044318,
+ "mean_feedback_forward_relative_error": 0.017440455328477055,
+ "min_feedback_forward_cosine": 0.9997293949127197
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 60605,
+ "train_examples": 50000,
+ "train_loss": 0.020610975865125655
+ },
+ {
+ "epoch": 156,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023287441581487656,
+ "mean_feedback_forward_cosine": 0.9998409842189989,
+ "mean_feedback_forward_relative_error": 0.01744029624387622,
+ "min_feedback_forward_cosine": 0.9997294545173645
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 60996,
+ "train_examples": 50000,
+ "train_loss": 0.01996522889852524
+ },
+ {
+ "epoch": 157,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023287393152713776,
+ "mean_feedback_forward_cosine": 0.9998410093156915,
+ "mean_feedback_forward_relative_error": 0.01744013637500374,
+ "min_feedback_forward_cosine": 0.9997293949127197
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 61387,
+ "train_examples": 50000,
+ "train_loss": 0.019278925200998782
+ },
+ {
+ "epoch": 158,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023287342861294746,
+ "mean_feedback_forward_cosine": 0.999840962259393,
+ "mean_feedback_forward_relative_error": 0.01743997809918303,
+ "min_feedback_forward_cosine": 0.9997293949127197
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 61778,
+ "train_examples": 50000,
+ "train_loss": 0.019621174844652416
+ },
+ {
+ "epoch": 159,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023287300020456314,
+ "mean_feedback_forward_cosine": 0.9998409904931721,
+ "mean_feedback_forward_relative_error": 0.01743981587749563,
+ "min_feedback_forward_cosine": 0.9997294545173645
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 62169,
+ "train_examples": 50000,
+ "train_loss": 0.019776837486624718
+ },
+ {
+ "epoch": 160,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023287255316972733,
+ "mean_feedback_forward_cosine": 0.9998409810819124,
+ "mean_feedback_forward_relative_error": 0.017439663826831077,
+ "min_feedback_forward_cosine": 0.9997294545173645
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 62560,
+ "train_examples": 50000,
+ "train_loss": 0.020378491938710212
+ },
+ {
+ "epoch": 161,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023287208750844002,
+ "mean_feedback_forward_cosine": 0.9998410187269512,
+ "mean_feedback_forward_relative_error": 0.01743950844301205,
+ "min_feedback_forward_cosine": 0.9997295141220093
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 62951,
+ "train_examples": 50000,
+ "train_loss": 0.0198153387182951
+ },
+ {
+ "epoch": 162,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02328716591000557,
+ "mean_feedback_forward_cosine": 0.9998409873560855,
+ "mean_feedback_forward_relative_error": 0.01743934813298677,
+ "min_feedback_forward_cosine": 0.9997293949127197
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 63342,
+ "train_examples": 50000,
+ "train_loss": 0.017996792468428613
+ },
+ {
+ "epoch": 163,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023287123069167137,
+ "mean_feedback_forward_cosine": 0.9998410061786049,
+ "mean_feedback_forward_relative_error": 0.01743918524957017,
+ "min_feedback_forward_cosine": 0.9997294545173645
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 63733,
+ "train_examples": 50000,
+ "train_loss": 0.01919635002851486
+ },
+ {
+ "epoch": 164,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023287080228328705,
+ "mean_feedback_forward_cosine": 0.9998409936302587,
+ "mean_feedback_forward_relative_error": 0.017439030993141626,
+ "min_feedback_forward_cosine": 0.9997294545173645
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 64124,
+ "train_examples": 50000,
+ "train_loss": 0.017838535489886998
+ },
+ {
+ "epoch": 165,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023287048563361168,
+ "mean_feedback_forward_cosine": 0.9998410061786049,
+ "mean_feedback_forward_relative_error": 0.017438872006574745,
+ "min_feedback_forward_cosine": 0.9997293949127197
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 64515,
+ "train_examples": 50000,
+ "train_loss": 0.017671099393367768
+ },
+ {
+ "epoch": 166,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023287005722522736,
+ "mean_feedback_forward_cosine": 0.9998410093156915,
+ "mean_feedback_forward_relative_error": 0.017438715593399184,
+ "min_feedback_forward_cosine": 0.999729573726654
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 64906,
+ "train_examples": 50000,
+ "train_loss": 0.01845243679612875
+ },
+ {
+ "epoch": 167,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023286962881684303,
+ "mean_feedback_forward_cosine": 0.9998410061786049,
+ "mean_feedback_forward_relative_error": 0.017438560944834824,
+ "min_feedback_forward_cosine": 0.9997294545173645
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 65297,
+ "train_examples": 50000,
+ "train_loss": 0.01822822848379612
+ },
+ {
+ "epoch": 168,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02328692562878132,
+ "mean_feedback_forward_cosine": 0.9998409999044318,
+ "mean_feedback_forward_relative_error": 0.01743840999705227,
+ "min_feedback_forward_cosine": 0.9997294545173645
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 65688,
+ "train_examples": 50000,
+ "train_loss": 0.018309657716155052
+ },
+ {
+ "epoch": 169,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023286882787942886,
+ "mean_feedback_forward_cosine": 0.9998410030415184,
+ "mean_feedback_forward_relative_error": 0.017438249956620366,
+ "min_feedback_forward_cosine": 0.9997294545173645
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 66079,
+ "train_examples": 50000,
+ "train_loss": 0.01790042732924223
+ },
+ {
+ "epoch": 170,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023286836221814156,
+ "mean_feedback_forward_cosine": 0.9998410093156915,
+ "mean_feedback_forward_relative_error": 0.01743809888629537,
+ "min_feedback_forward_cosine": 0.9997295141220093
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 66470,
+ "train_examples": 50000,
+ "train_loss": 0.01763336478382349
+ },
+ {
+ "epoch": 171,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023286795243620872,
+ "mean_feedback_forward_cosine": 0.999841012452778,
+ "mean_feedback_forward_relative_error": 0.017437945757257312,
+ "min_feedback_forward_cosine": 0.9997294545173645
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 66861,
+ "train_examples": 50000,
+ "train_loss": 0.017897946850806475
+ },
+ {
+ "epoch": 172,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023286746814846992,
+ "mean_feedback_forward_cosine": 0.9998410030415184,
+ "mean_feedback_forward_relative_error": 0.01743779211354099,
+ "min_feedback_forward_cosine": 0.9997295141220093
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 67252,
+ "train_examples": 50000,
+ "train_loss": 0.01726263719558716
+ },
+ {
+ "epoch": 173,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02328670583665371,
+ "mean_feedback_forward_cosine": 0.999841034412384,
+ "mean_feedback_forward_relative_error": 0.017437638665892576,
+ "min_feedback_forward_cosine": 0.9997294545173645
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 67643,
+ "train_examples": 50000,
+ "train_loss": 0.0170649010553956
+ },
+ {
+ "epoch": 174,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023286666721105576,
+ "mean_feedback_forward_cosine": 0.9998409936302587,
+ "mean_feedback_forward_relative_error": 0.01743748320854808,
+ "min_feedback_forward_cosine": 0.9997295141220093
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 68034,
+ "train_examples": 50000,
+ "train_loss": 0.017460184436738493
+ },
+ {
+ "epoch": 175,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023286623880267143,
+ "mean_feedback_forward_cosine": 0.9998409936302587,
+ "mean_feedback_forward_relative_error": 0.017437326574796123,
+ "min_feedback_forward_cosine": 0.9997295141220093
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 68425,
+ "train_examples": 50000,
+ "train_loss": 0.01830650190651417
+ },
+ {
+ "epoch": 176,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023286588490009308,
+ "mean_feedback_forward_cosine": 0.9998409967673453,
+ "mean_feedback_forward_relative_error": 0.017437181705118793,
+ "min_feedback_forward_cosine": 0.9997295141220093
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 68816,
+ "train_examples": 50000,
+ "train_loss": 0.017178862721920013
+ },
+ {
+ "epoch": 177,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023286549374461174,
+ "mean_feedback_forward_cosine": 0.9998409999044318,
+ "mean_feedback_forward_relative_error": 0.017437034556152003,
+ "min_feedback_forward_cosine": 0.9997294545173645
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 69207,
+ "train_examples": 50000,
+ "train_loss": 0.017217900459468364
+ },
+ {
+ "epoch": 178,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02328650653362274,
+ "mean_feedback_forward_cosine": 0.9998410312752974,
+ "mean_feedback_forward_relative_error": 0.017436881059486615,
+ "min_feedback_forward_cosine": 0.9997294545173645
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 69598,
+ "train_examples": 50000,
+ "train_loss": 0.017271476043015717
+ },
+ {
+ "epoch": 179,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02328646555542946,
+ "mean_feedback_forward_cosine": 0.9998410187269512,
+ "mean_feedback_forward_relative_error": 0.017436732734112365,
+ "min_feedback_forward_cosine": 0.9997295141220093
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 69989,
+ "train_examples": 50000,
+ "train_loss": 0.017563968722224235
+ },
+ {
+ "epoch": 180,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023286422714591026,
+ "mean_feedback_forward_cosine": 0.9998410061786049,
+ "mean_feedback_forward_relative_error": 0.0174365807079563,
+ "min_feedback_forward_cosine": 0.9997293949127197
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 70380,
+ "train_examples": 50000,
+ "train_loss": 0.017162835009098053
+ },
+ {
+ "epoch": 181,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023286376148462296,
+ "mean_feedback_forward_cosine": 0.9998410250011244,
+ "mean_feedback_forward_relative_error": 0.017436427971054064,
+ "min_feedback_forward_cosine": 0.999729335308075
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 70771,
+ "train_examples": 50000,
+ "train_loss": 0.01722058118015528
+ },
+ {
+ "epoch": 182,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02328633703291416,
+ "mean_feedback_forward_cosine": 0.999841034412384,
+ "mean_feedback_forward_relative_error": 0.01743627927805248,
+ "min_feedback_forward_cosine": 0.9997294545173645
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 71162,
+ "train_examples": 50000,
+ "train_loss": 0.016867592164874077
+ },
+ {
+ "epoch": 183,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02328629046678543,
+ "mean_feedback_forward_cosine": 0.9998410312752974,
+ "mean_feedback_forward_relative_error": 0.0174361257568786,
+ "min_feedback_forward_cosine": 0.9997294545173645
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 71553,
+ "train_examples": 50000,
+ "train_loss": 0.01651648246407509
+ },
+ {
+ "epoch": 184,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023286247625947,
+ "mean_feedback_forward_cosine": 0.9998410093156915,
+ "mean_feedback_forward_relative_error": 0.017435978754962746,
+ "min_feedback_forward_cosine": 0.9997294545173645
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 71944,
+ "train_examples": 50000,
+ "train_loss": 0.016739701369702816
+ },
+ {
+ "epoch": 185,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023286206647753716,
+ "mean_feedback_forward_cosine": 0.9998410595090765,
+ "mean_feedback_forward_relative_error": 0.0174358288855537,
+ "min_feedback_forward_cosine": 0.9997293949127197
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 72335,
+ "train_examples": 50000,
+ "train_loss": 0.016913250033557415
+ },
+ {
+ "epoch": 186,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02328617312014103,
+ "mean_feedback_forward_cosine": 0.9998410093156915,
+ "mean_feedback_forward_relative_error": 0.017435678305398477,
+ "min_feedback_forward_cosine": 0.9997295141220093
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 72726,
+ "train_examples": 50000,
+ "train_loss": 0.017181944202184676
+ },
+ {
+ "epoch": 187,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023286132141947746,
+ "mean_feedback_forward_cosine": 0.9998410312752974,
+ "mean_feedback_forward_relative_error": 0.01743552676941219,
+ "min_feedback_forward_cosine": 0.9997294545173645
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 73117,
+ "train_examples": 50000,
+ "train_loss": 0.017223249304294586
+ },
+ {
+ "epoch": 188,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023286093026399612,
+ "mean_feedback_forward_cosine": 0.9998410312752974,
+ "mean_feedback_forward_relative_error": 0.017435371581661075,
+ "min_feedback_forward_cosine": 0.9997295141220093
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 73508,
+ "train_examples": 50000,
+ "train_loss": 0.015878218078613282
+ },
+ {
+ "epoch": 189,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02328605018556118,
+ "mean_feedback_forward_cosine": 0.9998410281382109,
+ "mean_feedback_forward_relative_error": 0.017435220143708744,
+ "min_feedback_forward_cosine": 0.9997293949127197
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 73899,
+ "train_examples": 50000,
+ "train_loss": 0.01752918921351433
+ },
+ {
+ "epoch": 190,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023286007344722748,
+ "mean_feedback_forward_cosine": 0.9998410438236437,
+ "mean_feedback_forward_relative_error": 0.01743506696565371,
+ "min_feedback_forward_cosine": 0.9997295141220093
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 74290,
+ "train_examples": 50000,
+ "train_loss": 0.016888238548636435
+ },
+ {
+ "epoch": 191,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023285968229174614,
+ "mean_feedback_forward_cosine": 0.9998410250011244,
+ "mean_feedback_forward_relative_error": 0.017434921189162293,
+ "min_feedback_forward_cosine": 0.9997295141220093
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 74681,
+ "train_examples": 50000,
+ "train_loss": 0.017404314740002154
+ },
+ {
+ "epoch": 192,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023285934701561928,
+ "mean_feedback_forward_cosine": 0.9998410250011244,
+ "mean_feedback_forward_relative_error": 0.017434772790262575,
+ "min_feedback_forward_cosine": 0.9997295141220093
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 75072,
+ "train_examples": 50000,
+ "train_loss": 0.015363169437050819
+ },
+ {
+ "epoch": 193,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023285893723368645,
+ "mean_feedback_forward_cosine": 0.9998410532349035,
+ "mean_feedback_forward_relative_error": 0.01743463029790866,
+ "min_feedback_forward_cosine": 0.9997295141220093
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 75463,
+ "train_examples": 50000,
+ "train_loss": 0.016273371571302413
+ },
+ {
+ "epoch": 194,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023285843431949615,
+ "mean_feedback_forward_cosine": 0.9998410626461631,
+ "mean_feedback_forward_relative_error": 0.01743447430137741,
+ "min_feedback_forward_cosine": 0.9997295141220093
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 75854,
+ "train_examples": 50000,
+ "train_loss": 0.01754783113926649
+ },
+ {
+ "epoch": 195,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02328581176698208,
+ "mean_feedback_forward_cosine": 0.9998410469607303,
+ "mean_feedback_forward_relative_error": 0.017434322422272282,
+ "min_feedback_forward_cosine": 0.9997294545173645
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 76245,
+ "train_examples": 50000,
+ "train_loss": 0.015270344339311123
+ },
+ {
+ "epoch": 196,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023285770788788795,
+ "mean_feedback_forward_cosine": 0.9998410187269512,
+ "mean_feedback_forward_relative_error": 0.01743417880252788,
+ "min_feedback_forward_cosine": 0.9997294545173645
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 76636,
+ "train_examples": 50000,
+ "train_loss": 0.016540868909955025
+ },
+ {
+ "epoch": 197,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02328573167324066,
+ "mean_feedback_forward_cosine": 0.9998410438236437,
+ "mean_feedback_forward_relative_error": 0.017434029692881985,
+ "min_feedback_forward_cosine": 0.9997295141220093
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 77027,
+ "train_examples": 50000,
+ "train_loss": 0.01606601231575012
+ },
+ {
+ "epoch": 198,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023285694420337677,
+ "mean_feedback_forward_cosine": 0.9998410500978169,
+ "mean_feedback_forward_relative_error": 0.01743388271547462,
+ "min_feedback_forward_cosine": 0.9997294545173645
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 77418,
+ "train_examples": 50000,
+ "train_loss": 0.016473623100817205
+ },
+ {
+ "epoch": 199,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.023285653442144394,
+ "mean_feedback_forward_cosine": 0.9998410500978169,
+ "mean_feedback_forward_relative_error": 0.01743373333623535,
+ "min_feedback_forward_cosine": 0.9997295141220093
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 77809,
+ "train_examples": 50000,
+ "train_loss": 0.015586742779612542
+ },
+ {
+ "epoch": 200,
+ "feedback_tracking": {
+ "max_feedback_forward_relative_error": 0.02328561618924141,
+ "mean_feedback_forward_cosine": 0.9998410532349035,
+ "mean_feedback_forward_relative_error": 0.01743358883418535,
+ "min_feedback_forward_cosine": 0.9997295141220093
+ },
+ "lr": 0.0010000000000000002,
+ "output_lr": 0.0010000000000000002,
+ "step": 78200,
+ "train_examples": 50000,
+ "train_loss": 0.015274800401628017
+ }
+ ],
+ "evaluation_protocol": {
+ "test_evaluations": 1,
+ "test_used_for_selection": false,
+ "validation_evaluations": 0
+ },
+ "final": {
+ "accuracy": 0.9144,
+ "epoch": 200,
+ "evaluation_split": "test",
+ "finite": true,
+ "loss": 0.397072021484375,
+ "step": 78200
+ },
+ "hardware": {
+ "cuda_device_name": "NVIDIA GeForce GTX 1080",
+ "cuda_visible_devices": "7",
+ "device": "cuda",
+ "device_total_memory_bytes": 8507949056,
+ "peak_memory_allocated_bytes": 2162265600,
+ "peak_memory_reserved_bytes": 2864709632,
+ "torch_version": "2.3.1+cu118"
+ },
+ "protocol_family": "oral_a_cifar_local_resnet_development",
+ "provenance": {
+ "git_commit": "0008f2cd96c06bea98f43560867d93967c597711",
+ "git_tracked_dirty": false
+ },
+ "schema_version": 1,
+ "split": {
+ "cifar_source_files": [
+ {
+ "bytes": 31035704,
+ "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_1",
+ "sha256": "54636561a3ce25bd3e19253c6b0d8538147b0ae398331ac4a2d86c6d987368cd"
+ },
+ {
+ "bytes": 31035320,
+ "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_2",
+ "sha256": "766b2cef9fbc745cf056b3152224f7cf77163b330ea9a15f9392beb8b89bc5a8"
+ },
+ {
+ "bytes": 31035999,
+ "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_3",
+ "sha256": "0f00d98ebfb30b3ec0ad19f9756dc2630b89003e10525f5e148445e82aa6a1f9"
+ },
+ {
+ "bytes": 31035696,
+ "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_4",
+ "sha256": "3f7bb240661948b8f4d53e36ec720d8306f5668bd0071dcb4e6c947f78e9682b"
+ },
+ {
+ "bytes": 31035623,
+ "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_5",
+ "sha256": "d91802434d8376bbaeeadf58a737e3a1b12ac839077e931237e0dcd43adcb154"
+ },
+ {
+ "bytes": 31035526,
+ "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/test_batch",
+ "sha256": "f53d8d457504f7cff4ea9e021afcf0e0ad8e24a91f3fc42091b8adef61157831"
+ }
+ ],
+ "dataset": "cifar10",
+ "evaluation_split": "test",
+ "input_layout": "NCHW",
+ "input_shape": [
+ 3,
+ 32,
+ 32
+ ],
+ "loader_seed": 14,
+ "normalization_mean": [
+ 0.49140000343322754,
+ 0.4821999967098236,
+ 0.4465000033378601
+ ],
+ "normalization_std": [
+ 0.24699999392032623,
+ 0.2434999942779541,
+ 0.26159998774528503
+ ],
+ "split_from_training_only": true,
+ "split_seed": null,
+ "test_examples": 10000,
+ "train_examples": 50000,
+ "training_augmentation": "random_crop_32_padding_4_zero_then_horizontal_flip_p0.5",
+ "validation_class_counts": {},
+ "validation_examples": 0,
+ "validation_index_sha256": null
+ },
+ "timing": {
+ "apical_warmup_wall_s": 0.0,
+ "evaluation_wall_s": 0.7012748718261719,
+ "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": 6199.550989866257,
+ "train_wall_s": 6197.441554784775
+ },
+ "work": {
+ "apical_macs_per_example": 40108672,
+ "causal_scalar_observations": 0,
+ "components": {
+ "apical_projection_macs": 401086720000000,
+ "apical_regression_macs": 0,
+ "bp_reverse_macs_estimate": 0,
+ "kp_reciprocal_correlation_macs": 401086720000000,
+ "local_weight_correlation_macs": 405510400000000,
+ "mirror_feedback_prediction_macs": 0,
+ "mirror_local_correlation_macs": 0,
+ "mirror_response_macs": 0,
+ "ordinary_forward_macs": 405510400000000,
+ "perturbation_forward_macs": 0,
+ "warmup_clean_forward_macs": 0
+ },
+ "definition": "multiply-accumulates in conv/linear maps; one local weight correlation equals one forward-weight MAC count; BP reverse is estimated as one weight-gradient plus one activation-gradient convolution per forward convolution; mixed-traffic/predictor elementwise arithmetic is reported as a conservative operation estimate separately and is not folded into MACs",
+ "elementwise_operations_estimate": 0,
+ "forward_macs_per_example": 40551040,
+ "logical_batch_loss_queries": 0,
+ "mirror_probe_examples": 0,
+ "mirror_readout_probe_examples": 0,
+ "neutral_projection_observations": 0,
+ "per_example_cross_entropy_terms": 0,
+ "total_clean_forward_examples": 10000000,
+ "total_forward_equivalent_examples": 10000000,
+ "total_macs_estimate": 1613194240000000
+ }
+}