{
  "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": 13,
    "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/seed13_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": 13,
    "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": 0.9996645748615265,
    "feedback_forward_cosine": [
      0.9998794198036194,
      0.999871015548706,
      0.999824047088623,
      0.9998106956481934,
      0.9998115301132202,
      0.9997480511665344,
      0.9998006224632263,
      0.9998643398284912,
      0.9998229742050171,
      0.9997925758361816,
      0.9998612999916077,
      0.9998047947883606,
      0.9998359680175781,
      0.9999040961265564,
      0.9998941421508789,
      0.9998717308044434,
      0.9998804330825806,
      0.9997178316116333,
      0.9999887347221375
    ],
    "feedback_forward_norm_ratio": [
      0.9995253682136536,
      1.00080406665802,
      1.00015127658844,
      1.0006645917892456,
      0.9999712109565735,
      1.0010839700698853,
      0.9987130761146545,
      1.0003999471664429,
      0.9995747208595276,
      1.001042127609253,
      0.9997837543487549,
      1.00066339969635,
      0.999595046043396,
      1.0002201795578003,
      1.000038743019104,
      1.000614881515503,
      1.0001569986343384,
      1.0009127855300903,
      1.0022685527801514
    ],
    "feedback_forward_relative_error": [
      0.015535272657871246,
      0.016084633767604828,
      0.018762843683362007,
      0.019474836066365242,
      0.019413597881793976,
      0.022491998970508575,
      0.020000236108899117,
      0.016483338549733162,
      0.01881605200469494,
      0.020405489951372147,
      0.01665075123310089,
      0.019773559644818306,
      0.01811271347105503,
      0.013861091807484627,
      0.014551995322108269,
      0.01602592132985592,
      0.015464449301362038,
      0.023784108459949493,
      0.005254942923784256
    ],
    "innovation_negative_gradient_cosine": [
      0.999675452709198,
      0.9996551275253296,
      0.99966961145401,
      0.9996368288993835,
      0.9996786117553711,
      0.9996718168258667,
      0.9996957778930664,
      0.9996987581253052,
      0.9997268915176392,
      0.9997032284736633,
      0.999758780002594,
      0.9997425079345703,
      0.999778687953949,
      0.999828577041626,
      0.9998501539230347,
      0.9998707175254822,
      0.9999191761016846,
      0.9999624490737915,
      0.9106674790382385
    ],
    "mean_feedback_forward_cosine": 0.9998412791051363,
    "mean_feedback_forward_relative_error": 0.017418307007143374,
    "normalization_state": "training_batch_stats_without_running_update",
    "raw_negative_gradient_cosine": [
      0.999675452709198,
      0.9996551275253296,
      0.99966961145401,
      0.9996368288993835,
      0.9996786117553711,
      0.9996718168258667,
      0.9996957778930664,
      0.9996987581253052,
      0.9997268915176392,
      0.9997032284736633,
      0.999758780002594,
      0.9997425079345703,
      0.999778687953949,
      0.999828577041626,
      0.9998501539230347,
      0.9998707175254822,
      0.9999191761016846,
      0.9999624490737915,
      0.9106674790382385
    ],
    "teaching_negative_gradient_cosine": [
      0.999675452709198,
      0.9996551275253296,
      0.99966961145401,
      0.9996368288993835,
      0.9996786117553711,
      0.9996718168258667,
      0.9996957778930664,
      0.9996987581253052,
      0.9997268915176392,
      0.9997032284736633,
      0.999758780002594,
      0.9997425079345703,
      0.999778687953949,
      0.999828577041626,
      0.9998501539230347,
      0.9998707175254822,
      0.9999191761016846,
      0.9999624490737915,
      0.9106674790382385
    ],
    "wall_s": 0.09946918487548828
  },
  "epochs": [
    {
      "epoch": 1,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 1.3013068437576294,
        "mean_feedback_forward_cosine": 0.2745324773223777,
        "mean_feedback_forward_relative_error": 1.2014559193661338,
        "min_feedback_forward_cosine": 0.15550358593463898
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 391,
      "train_examples": 50000,
      "train_loss": 1.8105414065551757
    },
    {
      "epoch": 2,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 1.2381442785263062,
        "mean_feedback_forward_cosine": 0.40608432653703186,
        "mean_feedback_forward_relative_error": 1.0879248286548413,
        "min_feedback_forward_cosine": 0.24102555215358734
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 782,
      "train_examples": 50000,
      "train_loss": 1.3125261357879638
    },
    {
      "epoch": 3,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 1.194427728652954,
        "mean_feedback_forward_cosine": 0.5029313815267462,
        "mean_feedback_forward_relative_error": 0.9965660415197674,
        "min_feedback_forward_cosine": 0.2990334630012512
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 1173,
      "train_examples": 50000,
      "train_loss": 1.0516421655654906
    },
    {
      "epoch": 4,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 1.1473788022994995,
        "mean_feedback_forward_cosine": 0.5759214460849762,
        "mean_feedback_forward_relative_error": 0.9216395506733343,
        "min_feedback_forward_cosine": 0.35728707909584045
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 1564,
      "train_examples": 50000,
      "train_loss": 0.9011523639106751
    },
    {
      "epoch": 5,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 1.1042686700820923,
        "mean_feedback_forward_cosine": 0.6325716831182179,
        "mean_feedback_forward_relative_error": 0.8583083764502877,
        "min_feedback_forward_cosine": 0.4085603356361389
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 1955,
      "train_examples": 50000,
      "train_loss": 0.7793745107269288
    },
    {
      "epoch": 6,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 1.0634453296661377,
        "mean_feedback_forward_cosine": 0.6761405373874464,
        "mean_feedback_forward_relative_error": 0.8057038438947577,
        "min_feedback_forward_cosine": 0.453296422958374
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 2346,
      "train_examples": 50000,
      "train_loss": 0.7029642856788635
    },
    {
      "epoch": 7,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 1.0249955654144287,
        "mean_feedback_forward_cosine": 0.7115335527219271,
        "mean_feedback_forward_relative_error": 0.7601092598940197,
        "min_feedback_forward_cosine": 0.49332159757614136
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 2737,
      "train_examples": 50000,
      "train_loss": 0.6455783454704285
    },
    {
      "epoch": 8,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.9871819615364075,
        "mean_feedback_forward_cosine": 0.7421962932536477,
        "mean_feedback_forward_relative_error": 0.7184875638861405,
        "min_feedback_forward_cosine": 0.5310965776443481
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 3128,
      "train_examples": 50000,
      "train_loss": 0.603184449043274
    },
    {
      "epoch": 9,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.952755331993103,
        "mean_feedback_forward_cosine": 0.7669896514792192,
        "mean_feedback_forward_relative_error": 0.6828179014356512,
        "min_feedback_forward_cosine": 0.564058780670166
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 3519,
      "train_examples": 50000,
      "train_loss": 0.5623111936569214
    },
    {
      "epoch": 10,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.9190171957015991,
        "mean_feedback_forward_cosine": 0.7892683273867557,
        "mean_feedback_forward_relative_error": 0.6490171473277243,
        "min_feedback_forward_cosine": 0.5949069261550903
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 3910,
      "train_examples": 50000,
      "train_loss": 0.5330172318553924
    },
    {
      "epoch": 11,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.8857558965682983,
        "mean_feedback_forward_cosine": 0.8087137216015866,
        "mean_feedback_forward_relative_error": 0.6180857670934576,
        "min_feedback_forward_cosine": 0.6242749094963074
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 4301,
      "train_examples": 50000,
      "train_loss": 0.509774765663147
    },
    {
      "epoch": 12,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.8524502515792847,
        "mean_feedback_forward_cosine": 0.825825506135037,
        "mean_feedback_forward_relative_error": 0.589452138072566,
        "min_feedback_forward_cosine": 0.6520348787307739
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 4692,
      "train_examples": 50000,
      "train_loss": 0.4873462048435211
    },
    {
      "epoch": 13,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.8209549784660339,
        "mean_feedback_forward_cosine": 0.84129146839443,
        "mean_feedback_forward_relative_error": 0.5624135199346041,
        "min_feedback_forward_cosine": 0.6775722503662109
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 5083,
      "train_examples": 50000,
      "train_loss": 0.4644797011566162
    },
    {
      "epoch": 14,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.7899088263511658,
        "mean_feedback_forward_cosine": 0.8552724530822352,
        "mean_feedback_forward_relative_error": 0.5367354068316912,
        "min_feedback_forward_cosine": 0.7015966176986694
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 5474,
      "train_examples": 50000,
      "train_loss": 0.45303013045310975
    },
    {
      "epoch": 15,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.7603291273117065,
        "mean_feedback_forward_cosine": 0.8674388245532387,
        "mean_feedback_forward_relative_error": 0.5134045803233197,
        "min_feedback_forward_cosine": 0.7237777709960938
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 5865,
      "train_examples": 50000,
      "train_loss": 0.4340318734359741
    },
    {
      "epoch": 16,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.7316191792488098,
        "mean_feedback_forward_cosine": 0.8783629003324007,
        "mean_feedback_forward_relative_error": 0.4915046762478979,
        "min_feedback_forward_cosine": 0.7442220449447632
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 6256,
      "train_examples": 50000,
      "train_loss": 0.4222998009109497
    },
    {
      "epoch": 17,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.703751802444458,
        "mean_feedback_forward_cosine": 0.8885140607231542,
        "mean_feedback_forward_relative_error": 0.47033491024845525,
        "min_feedback_forward_cosine": 0.763218104839325
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 6647,
      "train_examples": 50000,
      "train_loss": 0.4074075182723999
    },
    {
      "epoch": 18,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.6776307225227356,
        "mean_feedback_forward_cosine": 0.897570760626542,
        "mean_feedback_forward_relative_error": 0.4505095058365872,
        "min_feedback_forward_cosine": 0.7804297804832458
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 7038,
      "train_examples": 50000,
      "train_loss": 0.3941944585990906
    },
    {
      "epoch": 19,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.6522223949432373,
        "mean_feedback_forward_cosine": 0.9056636032305265,
        "mean_feedback_forward_relative_error": 0.4321296324855403,
        "min_feedback_forward_cosine": 0.7966518402099609
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 7429,
      "train_examples": 50000,
      "train_loss": 0.3820874308586121
    },
    {
      "epoch": 20,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.6266651153564453,
        "mean_feedback_forward_cosine": 0.9131115298522147,
        "mean_feedback_forward_relative_error": 0.41454590464893143,
        "min_feedback_forward_cosine": 0.8121747374534607
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 7820,
      "train_examples": 50000,
      "train_loss": 0.3753539363288879
    },
    {
      "epoch": 21,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.6012489199638367,
        "mean_feedback_forward_cosine": 0.9200733335394609,
        "mean_feedback_forward_relative_error": 0.39740215398763357,
        "min_feedback_forward_cosine": 0.8270419239997864
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 8211,
      "train_examples": 50000,
      "train_loss": 0.3685603359031677
    },
    {
      "epoch": 22,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.5777385234832764,
        "mean_feedback_forward_cosine": 0.9262749552726746,
        "mean_feedback_forward_relative_error": 0.3814526161080913,
        "min_feedback_forward_cosine": 0.8401073217391968
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 8602,
      "train_examples": 50000,
      "train_loss": 0.35780021916389465
    },
    {
      "epoch": 23,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.5548236966133118,
        "mean_feedback_forward_cosine": 0.9320156731103596,
        "mean_feedback_forward_relative_error": 0.3660643430132615,
        "min_feedback_forward_cosine": 0.8525242209434509
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 8993,
      "train_examples": 50000,
      "train_loss": 0.3461114327335358
    },
    {
      "epoch": 24,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.532526969909668,
        "mean_feedback_forward_cosine": 0.9373489869268317,
        "mean_feedback_forward_relative_error": 0.3512582198569649,
        "min_feedback_forward_cosine": 0.8639460802078247
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 9384,
      "train_examples": 50000,
      "train_loss": 0.3433043381214142
    },
    {
      "epoch": 25,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.5109798908233643,
        "mean_feedback_forward_cosine": 0.9422720890296133,
        "mean_feedback_forward_relative_error": 0.3370238997434315,
        "min_feedback_forward_cosine": 0.8746147155761719
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 9775,
      "train_examples": 50000,
      "train_loss": 0.33923622669696807
    },
    {
      "epoch": 26,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.490651935338974,
        "mean_feedback_forward_cosine": 0.9466914754164847,
        "mean_feedback_forward_relative_error": 0.32364508666490255,
        "min_feedback_forward_cosine": 0.8841590881347656
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 10166,
      "train_examples": 50000,
      "train_loss": 0.326556473941803
    },
    {
      "epoch": 27,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.4708702266216278,
        "mean_feedback_forward_cosine": 0.9507664159724587,
        "mean_feedback_forward_relative_error": 0.31087132425684677,
        "min_feedback_forward_cosine": 0.8932213187217712
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 10557,
      "train_examples": 50000,
      "train_loss": 0.3248288626766205
    },
    {
      "epoch": 28,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.45137742161750793,
        "mean_feedback_forward_cosine": 0.9545714886564958,
        "mean_feedback_forward_relative_error": 0.2985292767223559,
        "min_feedback_forward_cosine": 0.9017136693000793
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 10948,
      "train_examples": 50000,
      "train_loss": 0.3213924436569214
    },
    {
      "epoch": 29,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.4335157871246338,
        "mean_feedback_forward_cosine": 0.9579601130987468,
        "mean_feedback_forward_relative_error": 0.28700707695986094,
        "min_feedback_forward_cosine": 0.909265398979187
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 11339,
      "train_examples": 50000,
      "train_loss": 0.3070706036186218
    },
    {
      "epoch": 30,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.4157837927341461,
        "mean_feedback_forward_cosine": 0.9612041994145042,
        "mean_feedback_forward_relative_error": 0.27560351318434667,
        "min_feedback_forward_cosine": 0.9164845943450928
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 11730,
      "train_examples": 50000,
      "train_loss": 0.3062127510166168
    },
    {
      "epoch": 31,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.3991200923919678,
        "mean_feedback_forward_cosine": 0.9641893160970587,
        "mean_feedback_forward_relative_error": 0.26467576348467875,
        "min_feedback_forward_cosine": 0.9229661226272583
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 12121,
      "train_examples": 50000,
      "train_loss": 0.3042983984851837
    },
    {
      "epoch": 32,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.38312971591949463,
        "mean_feedback_forward_cosine": 0.9668795905615154,
        "mean_feedback_forward_relative_error": 0.2544799458823706,
        "min_feedback_forward_cosine": 0.9289388060569763
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 12512,
      "train_examples": 50000,
      "train_loss": 0.2986329161453247
    },
    {
      "epoch": 33,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.36767882108688354,
        "mean_feedback_forward_cosine": 0.9693631874887567,
        "mean_feedback_forward_relative_error": 0.24468313941830083,
        "min_feedback_forward_cosine": 0.9344800710678101
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 12903,
      "train_examples": 50000,
      "train_loss": 0.29492065697669984
    },
    {
      "epoch": 34,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.3529456555843353,
        "mean_feedback_forward_cosine": 0.9716313857781259,
        "mean_feedback_forward_relative_error": 0.23538012018329219,
        "min_feedback_forward_cosine": 0.939528226852417
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 13294,
      "train_examples": 50000,
      "train_loss": 0.2894583631706238
    },
    {
      "epoch": 35,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.33894702792167664,
        "mean_feedback_forward_cosine": 0.9737380711655867,
        "mean_feedback_forward_relative_error": 0.22634300315066389,
        "min_feedback_forward_cosine": 0.9441406726837158
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 13685,
      "train_examples": 50000,
      "train_loss": 0.2836156642055511
    },
    {
      "epoch": 36,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.32576683163642883,
        "mean_feedback_forward_cosine": 0.9756966358736942,
        "mean_feedback_forward_relative_error": 0.21766045297447004,
        "min_feedback_forward_cosine": 0.9483542442321777
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 14076,
      "train_examples": 50000,
      "train_loss": 0.2780234604263306
    },
    {
      "epoch": 37,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.3122580349445343,
        "mean_feedback_forward_cosine": 0.9775765695069966,
        "mean_feedback_forward_relative_error": 0.20902353407520996,
        "min_feedback_forward_cosine": 0.9525107145309448
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 14467,
      "train_examples": 50000,
      "train_loss": 0.28286614376544955
    },
    {
      "epoch": 38,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.29952549934387207,
        "mean_feedback_forward_cosine": 0.9792511149456626,
        "mean_feedback_forward_relative_error": 0.20103395102839722,
        "min_feedback_forward_cosine": 0.9562575221061707
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 14858,
      "train_examples": 50000,
      "train_loss": 0.2785664082813263
    },
    {
      "epoch": 39,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.2874852120876312,
        "mean_feedback_forward_cosine": 0.9807931498477334,
        "mean_feedback_forward_relative_error": 0.1933269277215004,
        "min_feedback_forward_cosine": 0.9596430063247681
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 15249,
      "train_examples": 50000,
      "train_loss": 0.26943545625686643
    },
    {
      "epoch": 40,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.2761114537715912,
        "mean_feedback_forward_cosine": 0.9822197117303547,
        "mean_feedback_forward_relative_error": 0.18598330883603348,
        "min_feedback_forward_cosine": 0.9627286195755005
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 15640,
      "train_examples": 50000,
      "train_loss": 0.26704236145973204
    },
    {
      "epoch": 41,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.26505810022354126,
        "mean_feedback_forward_cosine": 0.9835320052347685,
        "mean_feedback_forward_relative_error": 0.17890466788881704,
        "min_feedback_forward_cosine": 0.96563720703125
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 16031,
      "train_examples": 50000,
      "train_loss": 0.26196546399116516
    },
    {
      "epoch": 42,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.25418269634246826,
        "mean_feedback_forward_cosine": 0.9847755055678519,
        "mean_feedback_forward_relative_error": 0.17198689242726878,
        "min_feedback_forward_cosine": 0.9683691263198853
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 16422,
      "train_examples": 50000,
      "train_loss": 0.26309292840957643
    },
    {
      "epoch": 43,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.24364709854125977,
        "mean_feedback_forward_cosine": 0.9859028521337008,
        "mean_feedback_forward_relative_error": 0.1654725741398962,
        "min_feedback_forward_cosine": 0.9709159135818481
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 16813,
      "train_examples": 50000,
      "train_loss": 0.26448858797073366
    },
    {
      "epoch": 44,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.23413823544979095,
        "mean_feedback_forward_cosine": 0.9869271046236942,
        "mean_feedback_forward_relative_error": 0.15927042694468246,
        "min_feedback_forward_cosine": 0.973129391670227
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 17204,
      "train_examples": 50000,
      "train_loss": 0.25197313119888304
    },
    {
      "epoch": 45,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.22454142570495605,
        "mean_feedback_forward_cosine": 0.9878971890399331,
        "mean_feedback_forward_relative_error": 0.1532468678135621,
        "min_feedback_forward_cosine": 0.9752538204193115
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 17595,
      "train_examples": 50000,
      "train_loss": 0.2589807509803772
    },
    {
      "epoch": 46,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.2152714878320694,
        "mean_feedback_forward_cosine": 0.9887983610755519,
        "mean_feedback_forward_relative_error": 0.14737906405015996,
        "min_feedback_forward_cosine": 0.9772320985794067
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 17986,
      "train_examples": 50000,
      "train_loss": 0.2515754291200638
    },
    {
      "epoch": 47,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.206338033080101,
        "mean_feedback_forward_cosine": 0.9896588262758756,
        "mean_feedback_forward_relative_error": 0.14157812022849134,
        "min_feedback_forward_cosine": 0.979076087474823
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 18377,
      "train_examples": 50000,
      "train_loss": 0.25133833992004395
    },
    {
      "epoch": 48,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.19791005551815033,
        "mean_feedback_forward_cosine": 0.9904338027301588,
        "mean_feedback_forward_relative_error": 0.13616145813935682,
        "min_feedback_forward_cosine": 0.9807275533676147
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 18768,
      "train_examples": 50000,
      "train_loss": 0.2520319424915314
    },
    {
      "epoch": 49,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.1899934709072113,
        "mean_feedback_forward_cosine": 0.9911384111956546,
        "mean_feedback_forward_relative_error": 0.13103488969959712,
        "min_feedback_forward_cosine": 0.9822343587875366
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 19159,
      "train_examples": 50000,
      "train_loss": 0.24880680119514464
    },
    {
      "epoch": 50,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.1821773797273636,
        "mean_feedback_forward_cosine": 0.9918174900506672,
        "mean_feedback_forward_relative_error": 0.12589611740488754,
        "min_feedback_forward_cosine": 0.9836522340774536
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 19550,
      "train_examples": 50000,
      "train_loss": 0.2517117203617096
    },
    {
      "epoch": 51,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.17499101161956787,
        "mean_feedback_forward_cosine": 0.9924169559227792,
        "mean_feedback_forward_relative_error": 0.12116963612405877,
        "min_feedback_forward_cosine": 0.9849077463150024
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 19941,
      "train_examples": 50000,
      "train_loss": 0.24075208876609802
    },
    {
      "epoch": 52,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.1675454080104828,
        "mean_feedback_forward_cosine": 0.993003741690987,
        "mean_feedback_forward_relative_error": 0.1163746302849368,
        "min_feedback_forward_cosine": 0.9861578941345215
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 20332,
      "train_examples": 50000,
      "train_loss": 0.2484998675251007
    },
    {
      "epoch": 53,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.16078555583953857,
        "mean_feedback_forward_cosine": 0.993527026552903,
        "mean_feedback_forward_relative_error": 0.11193676234075897,
        "min_feedback_forward_cosine": 0.987243115901947
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 20723,
      "train_examples": 50000,
      "train_loss": 0.24073161078453065
    },
    {
      "epoch": 54,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.1545904129743576,
        "mean_feedback_forward_cosine": 0.9939948038051003,
        "mean_feedback_forward_relative_error": 0.10778918156498357,
        "min_feedback_forward_cosine": 0.9882010817527771
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 21114,
      "train_examples": 50000,
      "train_loss": 0.23582501243591308
    },
    {
      "epoch": 55,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.1483989804983139,
        "mean_feedback_forward_cosine": 0.9944345982451188,
        "mean_feedback_forward_relative_error": 0.1037434255214114,
        "min_feedback_forward_cosine": 0.9891226291656494
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 21505,
      "train_examples": 50000,
      "train_loss": 0.23572792385101318
    },
    {
      "epoch": 56,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.14238110184669495,
        "mean_feedback_forward_cosine": 0.9948575026110599,
        "mean_feedback_forward_relative_error": 0.09972382728990756,
        "min_feedback_forward_cosine": 0.9899809956550598
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 21896,
      "train_examples": 50000,
      "train_loss": 0.23625793926239014
    },
    {
      "epoch": 57,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.13674940168857574,
        "mean_feedback_forward_cosine": 0.9952467711348283,
        "mean_feedback_forward_relative_error": 0.09586156649809134,
        "min_feedback_forward_cosine": 0.9907553195953369
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 22287,
      "train_examples": 50000,
      "train_loss": 0.2386486736679077
    },
    {
      "epoch": 58,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.1313994824886322,
        "mean_feedback_forward_cosine": 0.9955950065662986,
        "mean_feedback_forward_relative_error": 0.0922685687087084,
        "min_feedback_forward_cosine": 0.9914608001708984
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 22678,
      "train_examples": 50000,
      "train_loss": 0.23173666384220123
    },
    {
      "epoch": 59,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.12608899176120758,
        "mean_feedback_forward_cosine": 0.9959294827360856,
        "mean_feedback_forward_relative_error": 0.08868560783172909,
        "min_feedback_forward_cosine": 0.9921314716339111
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 23069,
      "train_examples": 50000,
      "train_loss": 0.23614755492687226
    },
    {
      "epoch": 60,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.12107746303081512,
        "mean_feedback_forward_cosine": 0.9962382975377535,
        "mean_feedback_forward_relative_error": 0.0852436446829846,
        "min_feedback_forward_cosine": 0.9927425384521484
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 23460,
      "train_examples": 50000,
      "train_loss": 0.23019294379234315
    },
    {
      "epoch": 61,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.11625780165195465,
        "mean_feedback_forward_cosine": 0.9965175578468725,
        "mean_feedback_forward_relative_error": 0.08200653113032642,
        "min_feedback_forward_cosine": 0.9933075308799744
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 23851,
      "train_examples": 50000,
      "train_loss": 0.22646534693717957
    },
    {
      "epoch": 62,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.11168581992387772,
        "mean_feedback_forward_cosine": 0.9967757432084334,
        "mean_feedback_forward_relative_error": 0.07890120031018007,
        "min_feedback_forward_cosine": 0.993818998336792
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 24242,
      "train_examples": 50000,
      "train_loss": 0.2278005755662918
    },
    {
      "epoch": 63,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.107177734375,
        "mean_feedback_forward_cosine": 0.9970227762272483,
        "mean_feedback_forward_relative_error": 0.0758151122810025,
        "min_feedback_forward_cosine": 0.994306206703186
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 24633,
      "train_examples": 50000,
      "train_loss": 0.22829332092285157
    },
    {
      "epoch": 64,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.10296737402677536,
        "mean_feedback_forward_cosine": 0.9972459642510665,
        "mean_feedback_forward_relative_error": 0.0729043290607239,
        "min_feedback_forward_cosine": 0.9947414994239807
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 25024,
      "train_examples": 50000,
      "train_loss": 0.22240068197250365
    },
    {
      "epoch": 65,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.09894931316375732,
        "mean_feedback_forward_cosine": 0.9974505117065028,
        "mean_feedback_forward_relative_error": 0.07013854619703795,
        "min_feedback_forward_cosine": 0.9951410889625549
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 25415,
      "train_examples": 50000,
      "train_loss": 0.22249927976608277
    },
    {
      "epoch": 66,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.09491506963968277,
        "mean_feedback_forward_cosine": 0.9976401423153124,
        "mean_feedback_forward_relative_error": 0.06747396937326382,
        "min_feedback_forward_cosine": 0.9955277442932129
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 25806,
      "train_examples": 50000,
      "train_loss": 0.22065757092952729
    },
    {
      "epoch": 67,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.09111560881137848,
        "mean_feedback_forward_cosine": 0.9978157313246476,
        "mean_feedback_forward_relative_error": 0.06491193851750147,
        "min_feedback_forward_cosine": 0.9958773851394653
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 26197,
      "train_examples": 50000,
      "train_loss": 0.22140870141983032
    },
    {
      "epoch": 68,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.08744850009679794,
        "mean_feedback_forward_cosine": 0.9979819435822336,
        "mean_feedback_forward_relative_error": 0.06239619762881806,
        "min_feedback_forward_cosine": 0.9962009191513062
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 26588,
      "train_examples": 50000,
      "train_loss": 0.22032229407310486
    },
    {
      "epoch": 69,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.08386296778917313,
        "mean_feedback_forward_cosine": 0.998133449177993,
        "mean_feedback_forward_relative_error": 0.06000860966742039,
        "min_feedback_forward_cosine": 0.996505856513977
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 26979,
      "train_examples": 50000,
      "train_loss": 0.22276923543930055
    },
    {
      "epoch": 70,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.08059180527925491,
        "mean_feedback_forward_cosine": 0.9982716315671017,
        "mean_feedback_forward_relative_error": 0.05773922389275149,
        "min_feedback_forward_cosine": 0.996772050857544
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 27370,
      "train_examples": 50000,
      "train_loss": 0.21753522283554078
    },
    {
      "epoch": 71,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.07729753106832504,
        "mean_feedback_forward_cosine": 0.9984038503546464,
        "mean_feedback_forward_relative_error": 0.05548090487718582,
        "min_feedback_forward_cosine": 0.9970303773880005
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 27761,
      "train_examples": 50000,
      "train_loss": 0.2184693816947937
    },
    {
      "epoch": 72,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.07408735901117325,
        "mean_feedback_forward_cosine": 0.9985291110841852,
        "mean_feedback_forward_relative_error": 0.05326454929615322,
        "min_feedback_forward_cosine": 0.9972708821296692
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 28152,
      "train_examples": 50000,
      "train_loss": 0.21877637727737426
    },
    {
      "epoch": 73,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.07115792483091354,
        "mean_feedback_forward_cosine": 0.9986389969524584,
        "mean_feedback_forward_relative_error": 0.05123646723988809,
        "min_feedback_forward_cosine": 0.9974827766418457
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 28543,
      "train_examples": 50000,
      "train_loss": 0.2199541541481018
    },
    {
      "epoch": 74,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.06825961172580719,
        "mean_feedback_forward_cosine": 0.9987425427687796,
        "mean_feedback_forward_relative_error": 0.04924720241443107,
        "min_feedback_forward_cosine": 0.9976828694343567
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 28934,
      "train_examples": 50000,
      "train_loss": 0.21874637095928193
    },
    {
      "epoch": 75,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.0656348392367363,
        "mean_feedback_forward_cosine": 0.9988354381762052,
        "mean_feedback_forward_relative_error": 0.047386938902108294,
        "min_feedback_forward_cosine": 0.9978573322296143
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 29325,
      "train_examples": 50000,
      "train_loss": 0.21468808176994325
    },
    {
      "epoch": 76,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.06299450248479843,
        "mean_feedback_forward_cosine": 0.9989244718300668,
        "mean_feedback_forward_relative_error": 0.04553594106906339,
        "min_feedback_forward_cosine": 0.9980255961418152
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 29716,
      "train_examples": 50000,
      "train_loss": 0.2171169550895691
    },
    {
      "epoch": 77,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.06045105680823326,
        "mean_feedback_forward_cosine": 0.9990069740696957,
        "mean_feedback_forward_relative_error": 0.043757025838682524,
        "min_feedback_forward_cosine": 0.998181939125061
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 30107,
      "train_examples": 50000,
      "train_loss": 0.2182661148452759
    },
    {
      "epoch": 78,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.05817895382642746,
        "mean_feedback_forward_cosine": 0.9990807207007157,
        "mean_feedback_forward_relative_error": 0.04209429937365808,
        "min_feedback_forward_cosine": 0.9983154535293579
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 30498,
      "train_examples": 50000,
      "train_loss": 0.21216947293281554
    },
    {
      "epoch": 79,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.055928610265254974,
        "mean_feedback_forward_cosine": 0.9991478292565596,
        "mean_feedback_forward_relative_error": 0.04052411313903959,
        "min_feedback_forward_cosine": 0.9984431862831116
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 30889,
      "train_examples": 50000,
      "train_loss": 0.21012849893569946
    },
    {
      "epoch": 80,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.05367901176214218,
        "mean_feedback_forward_cosine": 0.9992131057538485,
        "mean_feedback_forward_relative_error": 0.03893751631441869,
        "min_feedback_forward_cosine": 0.9985654354095459
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 31280,
      "train_examples": 50000,
      "train_loss": 0.2153410515832901
    },
    {
      "epoch": 81,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.0515558160841465,
        "mean_feedback_forward_cosine": 0.9992728484304327,
        "mean_feedback_forward_relative_error": 0.0374330644936938,
        "min_feedback_forward_cosine": 0.9986765384674072
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 31671,
      "train_examples": 50000,
      "train_loss": 0.21272299917697907
    },
    {
      "epoch": 82,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.04952530190348625,
        "mean_feedback_forward_cosine": 0.9993273866804022,
        "mean_feedback_forward_relative_error": 0.03600044876925255,
        "min_feedback_forward_cosine": 0.9987785816192627
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 32062,
      "train_examples": 50000,
      "train_loss": 0.20681360980033875
    },
    {
      "epoch": 83,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.0475892573595047,
        "mean_feedback_forward_cosine": 0.9993776302588614,
        "mean_feedback_forward_relative_error": 0.034629917948653825,
        "min_feedback_forward_cosine": 0.9988718628883362
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 32453,
      "train_examples": 50000,
      "train_loss": 0.20746429374694825
    },
    {
      "epoch": 84,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.04574531316757202,
        "mean_feedback_forward_cosine": 0.9994237517055712,
        "mean_feedback_forward_relative_error": 0.033321863089345004,
        "min_feedback_forward_cosine": 0.9989575743675232
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 32844,
      "train_examples": 50000,
      "train_loss": 0.20582803499698638
    },
    {
      "epoch": 85,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.04398082196712494,
        "mean_feedback_forward_cosine": 0.9994664192199707,
        "mean_feedback_forward_relative_error": 0.0320633217986477,
        "min_feedback_forward_cosine": 0.9990363717079163
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 33235,
      "train_examples": 50000,
      "train_loss": 0.2062230680656433
    },
    {
      "epoch": 86,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.04222811385989189,
        "mean_feedback_forward_cosine": 0.9995071511519583,
        "mean_feedback_forward_relative_error": 0.030812954088967098,
        "min_feedback_forward_cosine": 0.9991114139556885
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 33626,
      "train_examples": 50000,
      "train_loss": 0.20691422386169434
    },
    {
      "epoch": 87,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.040574077516794205,
        "mean_feedback_forward_cosine": 0.9995435946866086,
        "mean_feedback_forward_relative_error": 0.029650764784922724,
        "min_feedback_forward_cosine": 0.9991796016693115
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 34017,
      "train_examples": 50000,
      "train_loss": 0.20205317479610443
    },
    {
      "epoch": 88,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.0389895960688591,
        "mean_feedback_forward_cosine": 0.9995780053891634,
        "mean_feedback_forward_relative_error": 0.028509573030628656,
        "min_feedback_forward_cosine": 0.9992421865463257
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 34408,
      "train_examples": 50000,
      "train_loss": 0.2011510366487503
    },
    {
      "epoch": 89,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.03738435357809067,
        "mean_feedback_forward_cosine": 0.9996104428642675,
        "mean_feedback_forward_relative_error": 0.027394547715390984,
        "min_feedback_forward_cosine": 0.9993032217025757
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 34799,
      "train_examples": 50000,
      "train_loss": 0.20494672033786773
    },
    {
      "epoch": 90,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.03589176759123802,
        "mean_feedback_forward_cosine": 0.9996400412760282,
        "mean_feedback_forward_relative_error": 0.026334630913640324,
        "min_feedback_forward_cosine": 0.9993577599525452
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 35190,
      "train_examples": 50000,
      "train_loss": 0.20535221662521363
    },
    {
      "epoch": 91,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.03456330671906471,
        "mean_feedback_forward_cosine": 0.9996665590687802,
        "mean_feedback_forward_relative_error": 0.025342908932974462,
        "min_feedback_forward_cosine": 0.9994043111801147
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 35581,
      "train_examples": 50000,
      "train_loss": 0.19797568843841554
    },
    {
      "epoch": 92,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.03314723074436188,
        "mean_feedback_forward_cosine": 0.9996924274846127,
        "mean_feedback_forward_relative_error": 0.024340616686171608,
        "min_feedback_forward_cosine": 0.9994521141052246
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 35972,
      "train_examples": 50000,
      "train_loss": 0.19780106279850007
    },
    {
      "epoch": 93,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.03182428330183029,
        "mean_feedback_forward_cosine": 0.9997163540438602,
        "mean_feedback_forward_relative_error": 0.02337651140987873,
        "min_feedback_forward_cosine": 0.9994949102401733
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 36363,
      "train_examples": 50000,
      "train_loss": 0.20516345916748047
    },
    {
      "epoch": 94,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.03060869127511978,
        "mean_feedback_forward_cosine": 0.9997371309681943,
        "mean_feedback_forward_relative_error": 0.02250075752013608,
        "min_feedback_forward_cosine": 0.9995328187942505
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 36754,
      "train_examples": 50000,
      "train_loss": 0.19811729461669922
    },
    {
      "epoch": 95,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.029402973130345345,
        "mean_feedback_forward_cosine": 0.9997569228473463,
        "mean_feedback_forward_relative_error": 0.021640330356986898,
        "min_feedback_forward_cosine": 0.9995688796043396
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 37145,
      "train_examples": 50000,
      "train_loss": 0.20136297669410705
    },
    {
      "epoch": 96,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.028252501040697098,
        "mean_feedback_forward_cosine": 0.9997750489335311,
        "mean_feedback_forward_relative_error": 0.020818151582620646,
        "min_feedback_forward_cosine": 0.9996020793914795
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 37536,
      "train_examples": 50000,
      "train_loss": 0.19526903770446777
    },
    {
      "epoch": 97,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02711074985563755,
        "mean_feedback_forward_cosine": 0.9997925350540563,
        "mean_feedback_forward_relative_error": 0.019993915663738,
        "min_feedback_forward_cosine": 0.9996333718299866
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 37927,
      "train_examples": 50000,
      "train_loss": 0.19873701023101806
    },
    {
      "epoch": 98,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.026029979810118675,
        "mean_feedback_forward_cosine": 0.9998086283081457,
        "mean_feedback_forward_relative_error": 0.019204675097410615,
        "min_feedback_forward_cosine": 0.9996620416641235
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 38318,
      "train_examples": 50000,
      "train_loss": 0.20269780053138733
    },
    {
      "epoch": 99,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.025008725002408028,
        "mean_feedback_forward_cosine": 0.9998231498818648,
        "mean_feedback_forward_relative_error": 0.018463870878086278,
        "min_feedback_forward_cosine": 0.9996881484985352
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 38709,
      "train_examples": 50000,
      "train_loss": 0.19956941916942597
    },
    {
      "epoch": 100,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02405368536710739,
        "mean_feedback_forward_cosine": 0.9998362064361572,
        "mean_feedback_forward_relative_error": 0.017763595611445214,
        "min_feedback_forward_cosine": 0.9997115135192871
      },
      "lr": 0.1,
      "output_lr": 0.1,
      "step": 39100,
      "train_examples": 50000,
      "train_loss": 0.19267467247486114
    },
    {
      "epoch": 101,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.024044588208198547,
        "mean_feedback_forward_cosine": 0.9998363727017453,
        "mean_feedback_forward_relative_error": 0.01775366653639235,
        "min_feedback_forward_cosine": 0.9997116327285767
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 39491,
      "train_examples": 50000,
      "train_loss": 0.1250617404985428
    },
    {
      "epoch": 102,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.024039901793003082,
        "mean_feedback_forward_cosine": 0.9998364417176497,
        "mean_feedback_forward_relative_error": 0.01774644177701128,
        "min_feedback_forward_cosine": 0.9997117519378662
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 39882,
      "train_examples": 50000,
      "train_loss": 0.09932032520771027
    },
    {
      "epoch": 103,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02403544820845127,
        "mean_feedback_forward_cosine": 0.9998365389673334,
        "mean_feedback_forward_relative_error": 0.017739429922872467,
        "min_feedback_forward_cosine": 0.9997118711471558
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 40273,
      "train_examples": 50000,
      "train_loss": 0.08672679335355758
    },
    {
      "epoch": 104,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.024031005799770355,
        "mean_feedback_forward_cosine": 0.9998366299428438,
        "mean_feedback_forward_relative_error": 0.017732631880789995,
        "min_feedback_forward_cosine": 0.9997119903564453
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 40664,
      "train_examples": 50000,
      "train_loss": 0.08061454909801483
    },
    {
      "epoch": 105,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02402658201754093,
        "mean_feedback_forward_cosine": 0.9998367177812677,
        "mean_feedback_forward_relative_error": 0.01772601821606881,
        "min_feedback_forward_cosine": 0.9997119903564453
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 41055,
      "train_examples": 50000,
      "train_loss": 0.07662240268230439
    },
    {
      "epoch": 106,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.024022245779633522,
        "mean_feedback_forward_cosine": 0.9998367993455184,
        "mean_feedback_forward_relative_error": 0.01771955042587299,
        "min_feedback_forward_cosine": 0.9997121691703796
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 41446,
      "train_examples": 50000,
      "train_loss": 0.0723095839548111
    },
    {
      "epoch": 107,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02401798777282238,
        "mean_feedback_forward_cosine": 0.9998368714985094,
        "mean_feedback_forward_relative_error": 0.017713176797291164,
        "min_feedback_forward_cosine": 0.999712347984314
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 41837,
      "train_examples": 50000,
      "train_loss": 0.06760056983351707
    },
    {
      "epoch": 108,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02401360310614109,
        "mean_feedback_forward_cosine": 0.9998369593369333,
        "mean_feedback_forward_relative_error": 0.017707002545265776,
        "min_feedback_forward_cosine": 0.999712347984314
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 42228,
      "train_examples": 50000,
      "train_loss": 0.0668923641860485
    },
    {
      "epoch": 109,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02400944195687771,
        "mean_feedback_forward_cosine": 0.9998370314899244,
        "mean_feedback_forward_relative_error": 0.01770084191996016,
        "min_feedback_forward_cosine": 0.9997124671936035
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 42619,
      "train_examples": 50000,
      "train_loss": 0.06146789957761765
    },
    {
      "epoch": 110,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.024005111306905746,
        "mean_feedback_forward_cosine": 0.9998370910945692,
        "mean_feedback_forward_relative_error": 0.017694778201219282,
        "min_feedback_forward_cosine": 0.9997125267982483
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 43010,
      "train_examples": 50000,
      "train_loss": 0.060262521543502805
    },
    {
      "epoch": 111,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.024000661447644234,
        "mean_feedback_forward_cosine": 0.9998371726588199,
        "mean_feedback_forward_relative_error": 0.017688673822895476,
        "min_feedback_forward_cosine": 0.9997127056121826
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 43401,
      "train_examples": 50000,
      "train_loss": 0.057219693999290465
    },
    {
      "epoch": 112,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023996371775865555,
        "mean_feedback_forward_cosine": 0.9998372510859841,
        "mean_feedback_forward_relative_error": 0.01768273159273361,
        "min_feedback_forward_cosine": 0.9997128844261169
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 43792,
      "train_examples": 50000,
      "train_loss": 0.053790216858386995
    },
    {
      "epoch": 113,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023991623893380165,
        "mean_feedback_forward_cosine": 0.999837360884014,
        "mean_feedback_forward_relative_error": 0.0176766182443029,
        "min_feedback_forward_cosine": 0.9997128844261169
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 44183,
      "train_examples": 50000,
      "train_loss": 0.057131857721805575
    },
    {
      "epoch": 114,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023986918851733208,
        "mean_feedback_forward_cosine": 0.9998374393111781,
        "mean_feedback_forward_relative_error": 0.017670371814777975,
        "min_feedback_forward_cosine": 0.9997129440307617
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 44574,
      "train_examples": 50000,
      "train_loss": 0.05416111909151077
    },
    {
      "epoch": 115,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023982137441635132,
        "mean_feedback_forward_cosine": 0.9998375302866885,
        "mean_feedback_forward_relative_error": 0.01766435196623206,
        "min_feedback_forward_cosine": 0.9997131824493408
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 44965,
      "train_examples": 50000,
      "train_loss": 0.0513282914686203
    },
    {
      "epoch": 116,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023977575823664665,
        "mean_feedback_forward_cosine": 0.9998375993025931,
        "mean_feedback_forward_relative_error": 0.017658326603276164,
        "min_feedback_forward_cosine": 0.9997131824493408
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 45356,
      "train_examples": 50000,
      "train_loss": 0.0487374546456337
    },
    {
      "epoch": 117,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02397250384092331,
        "mean_feedback_forward_cosine": 0.9998376808668438,
        "mean_feedback_forward_relative_error": 0.017652023265040236,
        "min_feedback_forward_cosine": 0.9997134208679199
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 45747,
      "train_examples": 50000,
      "train_loss": 0.04952977873086929
    },
    {
      "epoch": 118,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02396773174405098,
        "mean_feedback_forward_cosine": 0.9998377812536139,
        "mean_feedback_forward_relative_error": 0.017645936851438723,
        "min_feedback_forward_cosine": 0.9997135400772095
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 46138,
      "train_examples": 50000,
      "train_loss": 0.047488643206357956
    },
    {
      "epoch": 119,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02396288327872753,
        "mean_feedback_forward_cosine": 0.9998378722291243,
        "mean_feedback_forward_relative_error": 0.017639746962997475,
        "min_feedback_forward_cosine": 0.999713659286499
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 46529,
      "train_examples": 50000,
      "train_loss": 0.045404925146102906
    },
    {
      "epoch": 120,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02395789511501789,
        "mean_feedback_forward_cosine": 0.9998379412450289,
        "mean_feedback_forward_relative_error": 0.017633568005342232,
        "min_feedback_forward_cosine": 0.9997137188911438
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 46920,
      "train_examples": 50000,
      "train_loss": 0.044493886718153956
    },
    {
      "epoch": 121,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023953141644597054,
        "mean_feedback_forward_cosine": 0.999838041631799,
        "mean_feedback_forward_relative_error": 0.017627432942390442,
        "min_feedback_forward_cosine": 0.9997138381004333
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 47311,
      "train_examples": 50000,
      "train_loss": 0.044326152799129485
    },
    {
      "epoch": 122,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02394844964146614,
        "mean_feedback_forward_cosine": 0.9998381388814825,
        "mean_feedback_forward_relative_error": 0.017621343146617476,
        "min_feedback_forward_cosine": 0.9997138977050781
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 47702,
      "train_examples": 50000,
      "train_loss": 0.04166632812976837
    },
    {
      "epoch": 123,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023942919448018074,
        "mean_feedback_forward_cosine": 0.9998382392682528,
        "mean_feedback_forward_relative_error": 0.017614795164646285,
        "min_feedback_forward_cosine": 0.9997140765190125
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 48093,
      "train_examples": 50000,
      "train_loss": 0.04328630323410034
    },
    {
      "epoch": 124,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023937642574310303,
        "mean_feedback_forward_cosine": 0.9998383051470706,
        "mean_feedback_forward_relative_error": 0.01760836448030252,
        "min_feedback_forward_cosine": 0.9997142553329468
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 48484,
      "train_examples": 50000,
      "train_loss": 0.0424068547603488
    },
    {
      "epoch": 125,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023933056741952896,
        "mean_feedback_forward_cosine": 0.9998384118080139,
        "mean_feedback_forward_relative_error": 0.01760227644914075,
        "min_feedback_forward_cosine": 0.9997143149375916
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 48875,
      "train_examples": 50000,
      "train_loss": 0.038611125857830046
    },
    {
      "epoch": 126,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02392778918147087,
        "mean_feedback_forward_cosine": 0.9998385153318706,
        "mean_feedback_forward_relative_error": 0.017595812310709766,
        "min_feedback_forward_cosine": 0.9997144341468811
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 49266,
      "train_examples": 50000,
      "train_loss": 0.03885844133496284
    },
    {
      "epoch": 127,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023923058062791824,
        "mean_feedback_forward_cosine": 0.9998386094444677,
        "mean_feedback_forward_relative_error": 0.017589850253180453,
        "min_feedback_forward_cosine": 0.9997144937515259
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 49657,
      "train_examples": 50000,
      "train_loss": 0.035599832309484485
    },
    {
      "epoch": 128,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02391800470650196,
        "mean_feedback_forward_cosine": 0.999838694145805,
        "mean_feedback_forward_relative_error": 0.017583573506654876,
        "min_feedback_forward_cosine": 0.9997146129608154
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 50048,
      "train_examples": 50000,
      "train_loss": 0.03567034682750702
    },
    {
      "epoch": 129,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023912601172924042,
        "mean_feedback_forward_cosine": 0.999838788258402,
        "mean_feedback_forward_relative_error": 0.017577025034513912,
        "min_feedback_forward_cosine": 0.9997147917747498
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 50439,
      "train_examples": 50000,
      "train_loss": 0.036537800406217576
    },
    {
      "epoch": 130,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023907097056508064,
        "mean_feedback_forward_cosine": 0.9998389137418646,
        "mean_feedback_forward_relative_error": 0.01757047837600112,
        "min_feedback_forward_cosine": 0.9997149705886841
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 50830,
      "train_examples": 50000,
      "train_loss": 0.03574167158961296
    },
    {
      "epoch": 131,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02390183135867119,
        "mean_feedback_forward_cosine": 0.9998389890319422,
        "mean_feedback_forward_relative_error": 0.01756406957773786,
        "min_feedback_forward_cosine": 0.9997150301933289
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 51221,
      "train_examples": 50000,
      "train_loss": 0.03439360812664032
    },
    {
      "epoch": 132,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023896407335996628,
        "mean_feedback_forward_cosine": 0.9998391302008378,
        "mean_feedback_forward_relative_error": 0.017557448805554918,
        "min_feedback_forward_cosine": 0.999715268611908
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 51612,
      "train_examples": 50000,
      "train_loss": 0.03394215288341045
    },
    {
      "epoch": 133,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023890849202871323,
        "mean_feedback_forward_cosine": 0.9998392117650885,
        "mean_feedback_forward_relative_error": 0.01755088256475957,
        "min_feedback_forward_cosine": 0.999715268611908
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 52003,
      "train_examples": 50000,
      "train_loss": 0.03237517976492643
    },
    {
      "epoch": 134,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02388530597090721,
        "mean_feedback_forward_cosine": 0.9998392933293393,
        "mean_feedback_forward_relative_error": 0.017544142828371963,
        "min_feedback_forward_cosine": 0.9997154474258423
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 52394,
      "train_examples": 50000,
      "train_loss": 0.03302400237858295
    },
    {
      "epoch": 135,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023880058899521828,
        "mean_feedback_forward_cosine": 0.9998394219498885,
        "mean_feedback_forward_relative_error": 0.01753760462528781,
        "min_feedback_forward_cosine": 0.9997156262397766
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 52785,
      "train_examples": 50000,
      "train_loss": 0.03138346595227719
    },
    {
      "epoch": 136,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023874498903751373,
        "mean_feedback_forward_cosine": 0.9998395035141393,
        "mean_feedback_forward_relative_error": 0.017530909396315877,
        "min_feedback_forward_cosine": 0.9997156858444214
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 53176,
      "train_examples": 50000,
      "train_loss": 0.03181140359580517
    },
    {
      "epoch": 137,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02386857010424137,
        "mean_feedback_forward_cosine": 0.9998396321346885,
        "mean_feedback_forward_relative_error": 0.017523922908462976,
        "min_feedback_forward_cosine": 0.9997158646583557
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 53567,
      "train_examples": 50000,
      "train_loss": 0.03216016135275364
    },
    {
      "epoch": 138,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02386280708014965,
        "mean_feedback_forward_cosine": 0.9998397293843722,
        "mean_feedback_forward_relative_error": 0.017516972350054665,
        "min_feedback_forward_cosine": 0.99971604347229
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 53958,
      "train_examples": 50000,
      "train_loss": 0.030717867047190667
    },
    {
      "epoch": 139,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023856565356254578,
        "mean_feedback_forward_cosine": 0.9998398360453153,
        "mean_feedback_forward_relative_error": 0.017509598922180503,
        "min_feedback_forward_cosine": 0.9997161030769348
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 54349,
      "train_examples": 50000,
      "train_loss": 0.03143452783703804
    },
    {
      "epoch": 140,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02385035715997219,
        "mean_feedback_forward_cosine": 0.9998399427062586,
        "mean_feedback_forward_relative_error": 0.01750256920135335,
        "min_feedback_forward_cosine": 0.9997162818908691
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 54740,
      "train_examples": 50000,
      "train_loss": 0.030715213534832
    },
    {
      "epoch": 141,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023844636976718903,
        "mean_feedback_forward_cosine": 0.9998400838751542,
        "mean_feedback_forward_relative_error": 0.01749557940485446,
        "min_feedback_forward_cosine": 0.9997165203094482
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 55131,
      "train_examples": 50000,
      "train_loss": 0.02935562146782875
    },
    {
      "epoch": 142,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023838603869080544,
        "mean_feedback_forward_cosine": 0.9998401842619243,
        "mean_feedback_forward_relative_error": 0.017488385207558934,
        "min_feedback_forward_cosine": 0.9997165203094482
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 55522,
      "train_examples": 50000,
      "train_loss": 0.029773226491212845
    },
    {
      "epoch": 143,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02383241057395935,
        "mean_feedback_forward_cosine": 0.9998403254308199,
        "mean_feedback_forward_relative_error": 0.01748096582626826,
        "min_feedback_forward_cosine": 0.9997167587280273
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 55913,
      "train_examples": 50000,
      "train_loss": 0.029120073041915894
    },
    {
      "epoch": 144,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02382666990160942,
        "mean_feedback_forward_cosine": 0.9998404509142825,
        "mean_feedback_forward_relative_error": 0.01747380638200986,
        "min_feedback_forward_cosine": 0.9997168779373169
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 56304,
      "train_examples": 50000,
      "train_loss": 0.02774130385518074
    },
    {
      "epoch": 145,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023820001631975174,
        "mean_feedback_forward_cosine": 0.9998405669864855,
        "mean_feedback_forward_relative_error": 0.017466036700888685,
        "min_feedback_forward_cosine": 0.9997169971466064
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 56695,
      "train_examples": 50000,
      "train_loss": 0.02875920710802078
    },
    {
      "epoch": 146,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02381327375769615,
        "mean_feedback_forward_cosine": 0.9998406736474288,
        "mean_feedback_forward_relative_error": 0.01745791073986574,
        "min_feedback_forward_cosine": 0.9997172355651855
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 57086,
      "train_examples": 50000,
      "train_loss": 0.029547973494529724
    },
    {
      "epoch": 147,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.0238062534481287,
        "mean_feedback_forward_cosine": 0.9998408273646706,
        "mean_feedback_forward_relative_error": 0.01744976073601528,
        "min_feedback_forward_cosine": 0.9997172355651855
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 57477,
      "train_examples": 50000,
      "train_loss": 0.029193686606884
    },
    {
      "epoch": 148,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023799123242497444,
        "mean_feedback_forward_cosine": 0.9998409653964796,
        "mean_feedback_forward_relative_error": 0.017441571469565754,
        "min_feedback_forward_cosine": 0.9997174143791199
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 57868,
      "train_examples": 50000,
      "train_loss": 0.029274324528574945
    },
    {
      "epoch": 149,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023792922496795654,
        "mean_feedback_forward_cosine": 0.9998411097024617,
        "mean_feedback_forward_relative_error": 0.017433960799519952,
        "min_feedback_forward_cosine": 0.999717652797699
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 58259,
      "train_examples": 50000,
      "train_loss": 0.02635092706412077
    },
    {
      "epoch": 150,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02378627471625805,
        "mean_feedback_forward_cosine": 0.9998411755812796,
        "mean_feedback_forward_relative_error": 0.017425957454466505,
        "min_feedback_forward_cosine": 0.9997177124023438
      },
      "lr": 0.010000000000000002,
      "output_lr": 0.010000000000000002,
      "step": 58650,
      "train_examples": 50000,
      "train_loss": 0.02678735072135925
    },
    {
      "epoch": 151,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023786203935742378,
        "mean_feedback_forward_cosine": 0.9998412100892318,
        "mean_feedback_forward_relative_error": 0.01742577704748041,
        "min_feedback_forward_cosine": 0.9997178912162781
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 59041,
      "train_examples": 50000,
      "train_loss": 0.02437458291053772
    },
    {
      "epoch": 152,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023786155506968498,
        "mean_feedback_forward_cosine": 0.9998411975408855,
        "mean_feedback_forward_relative_error": 0.01742561279158843,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 59432,
      "train_examples": 50000,
      "train_loss": 0.021376969789266585
    },
    {
      "epoch": 153,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02378610335290432,
        "mean_feedback_forward_cosine": 0.999841222637578,
        "mean_feedback_forward_relative_error": 0.017425447972001212,
        "min_feedback_forward_cosine": 0.9997177124023438
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 59823,
      "train_examples": 50000,
      "train_loss": 0.020672446417212487
    },
    {
      "epoch": 154,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023786064237356186,
        "mean_feedback_forward_cosine": 0.999841216363405,
        "mean_feedback_forward_relative_error": 0.01742528391217715,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 60214,
      "train_examples": 50000,
      "train_loss": 0.019626799536943434
    },
    {
      "epoch": 155,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023786017671227455,
        "mean_feedback_forward_cosine": 0.9998412132263184,
        "mean_feedback_forward_relative_error": 0.01742512281788023,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 60605,
      "train_examples": 50000,
      "train_loss": 0.020080795953273774
    },
    {
      "epoch": 156,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023785961791872978,
        "mean_feedback_forward_cosine": 0.9998412132263184,
        "mean_feedback_forward_relative_error": 0.017424962091210642,
        "min_feedback_forward_cosine": 0.9997177124023438
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 60996,
      "train_examples": 50000,
      "train_loss": 0.02060788645744324
    },
    {
      "epoch": 157,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023785918951034546,
        "mean_feedback_forward_cosine": 0.999841222637578,
        "mean_feedback_forward_relative_error": 0.017424809329799916,
        "min_feedback_forward_cosine": 0.9997177720069885
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 61387,
      "train_examples": 50000,
      "train_loss": 0.018866425810456278
    },
    {
      "epoch": 158,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023785877972841263,
        "mean_feedback_forward_cosine": 0.9998412195004915,
        "mean_feedback_forward_relative_error": 0.01742464568662016,
        "min_feedback_forward_cosine": 0.9997177124023438
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 61778,
      "train_examples": 50000,
      "train_loss": 0.018583618539571762
    },
    {
      "epoch": 159,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02378583699464798,
        "mean_feedback_forward_cosine": 0.9998412132263184,
        "mean_feedback_forward_relative_error": 0.01742448495995057,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 62169,
      "train_examples": 50000,
      "train_loss": 0.019112786593735217
    },
    {
      "epoch": 160,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023785792291164398,
        "mean_feedback_forward_cosine": 0.999841216363405,
        "mean_feedback_forward_relative_error": 0.01742433254165869,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 62560,
      "train_examples": 50000,
      "train_loss": 0.018837088904976845
    },
    {
      "epoch": 161,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023785745725035667,
        "mean_feedback_forward_cosine": 0.999841222637578,
        "mean_feedback_forward_relative_error": 0.017424173015905053,
        "min_feedback_forward_cosine": 0.9997177720069885
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 62951,
      "train_examples": 50000,
      "train_loss": 0.019263693317174913
    },
    {
      "epoch": 162,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023785704746842384,
        "mean_feedback_forward_cosine": 0.9998412038150587,
        "mean_feedback_forward_relative_error": 0.017424011406929868,
        "min_feedback_forward_cosine": 0.9997177124023438
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 63342,
      "train_examples": 50000,
      "train_loss": 0.018983551829755307
    },
    {
      "epoch": 163,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023785660043358803,
        "mean_feedback_forward_cosine": 0.9998412257746646,
        "mean_feedback_forward_relative_error": 0.017423859037654966,
        "min_feedback_forward_cosine": 0.9997177720069885
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 63733,
      "train_examples": 50000,
      "train_loss": 0.018922538774609567
    },
    {
      "epoch": 164,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023785622790455818,
        "mean_feedback_forward_cosine": 0.9998412289117512,
        "mean_feedback_forward_relative_error": 0.017423704144005712,
        "min_feedback_forward_cosine": 0.9997177124023438
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 64124,
      "train_examples": 50000,
      "train_loss": 0.018507695814073084
    },
    {
      "epoch": 165,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023785576224327087,
        "mean_feedback_forward_cosine": 0.999841244597184,
        "mean_feedback_forward_relative_error": 0.01742355297564676,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 64515,
      "train_examples": 50000,
      "train_loss": 0.01805218663930893
    },
    {
      "epoch": 166,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023785537108778954,
        "mean_feedback_forward_cosine": 0.9998412195004915,
        "mean_feedback_forward_relative_error": 0.01742339675853911,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 64906,
      "train_examples": 50000,
      "train_loss": 0.01841807530283928
    },
    {
      "epoch": 167,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023785486817359924,
        "mean_feedback_forward_cosine": 0.9998412289117512,
        "mean_feedback_forward_relative_error": 0.01742324448729816,
        "min_feedback_forward_cosine": 0.9997178912162781
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 65297,
      "train_examples": 50000,
      "train_loss": 0.018245168339163065
    },
    {
      "epoch": 168,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02378544770181179,
        "mean_feedback_forward_cosine": 0.9998412351859244,
        "mean_feedback_forward_relative_error": 0.017423090990632772,
        "min_feedback_forward_cosine": 0.9997177720069885
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 65688,
      "train_examples": 50000,
      "train_loss": 0.01727046649456024
    },
    {
      "epoch": 169,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02378540113568306,
        "mean_feedback_forward_cosine": 0.99984126655679,
        "mean_feedback_forward_relative_error": 0.017422934969593035,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 66079,
      "train_examples": 50000,
      "train_loss": 0.016953599601387977
    },
    {
      "epoch": 170,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023785360157489777,
        "mean_feedback_forward_cosine": 0.99984126655679,
        "mean_feedback_forward_relative_error": 0.017422790124424194,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 66470,
      "train_examples": 50000,
      "train_loss": 0.017145444607436656
    },
    {
      "epoch": 171,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023785313591361046,
        "mean_feedback_forward_cosine": 0.9998412100892318,
        "mean_feedback_forward_relative_error": 0.017422635206266454,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 66861,
      "train_examples": 50000,
      "train_loss": 0.017727336402088404
    },
    {
      "epoch": 172,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023785270750522614,
        "mean_feedback_forward_cosine": 0.9998412602826169,
        "mean_feedback_forward_relative_error": 0.017422484135941455,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 67252,
      "train_examples": 50000,
      "train_loss": 0.017764935253858565
    },
    {
      "epoch": 173,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023785226047039032,
        "mean_feedback_forward_cosine": 0.9998412571455303,
        "mean_feedback_forward_relative_error": 0.017422324414119908,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 67643,
      "train_examples": 50000,
      "train_loss": 0.017218405064195395
    },
    {
      "epoch": 174,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023785186931490898,
        "mean_feedback_forward_cosine": 0.9998412540084437,
        "mean_feedback_forward_relative_error": 0.017422170721386607,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 68034,
      "train_examples": 50000,
      "train_loss": 0.018007522631287576
    },
    {
      "epoch": 175,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023785151541233063,
        "mean_feedback_forward_cosine": 0.9998412696938765,
        "mean_feedback_forward_relative_error": 0.017422025851709277,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 68425,
      "train_examples": 50000,
      "train_loss": 0.017599506865143777
    },
    {
      "epoch": 176,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02378511242568493,
        "mean_feedback_forward_cosine": 0.9998412602826169,
        "mean_feedback_forward_relative_error": 0.01742187377653624,
        "min_feedback_forward_cosine": 0.9997178912162781
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 68816,
      "train_examples": 50000,
      "train_loss": 0.01672246292024851
    },
    {
      "epoch": 177,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023785067722201347,
        "mean_feedback_forward_cosine": 0.9998412351859244,
        "mean_feedback_forward_relative_error": 0.01742171569678344,
        "min_feedback_forward_cosine": 0.9997177720069885
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 69207,
      "train_examples": 50000,
      "train_loss": 0.017969189695864914
    },
    {
      "epoch": 178,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023785024881362915,
        "mean_feedback_forward_cosine": 0.9998412602826169,
        "mean_feedback_forward_relative_error": 0.017421570557512735,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 69598,
      "train_examples": 50000,
      "train_loss": 0.016756232603490354
    },
    {
      "epoch": 179,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023784982040524483,
        "mean_feedback_forward_cosine": 0.9998412540084437,
        "mean_feedback_forward_relative_error": 0.017421419781289603,
        "min_feedback_forward_cosine": 0.9997177720069885
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 69989,
      "train_examples": 50000,
      "train_loss": 0.01737038464665413
    },
    {
      "epoch": 180,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02378493919968605,
        "mean_feedback_forward_cosine": 0.9998412759680497,
        "mean_feedback_forward_relative_error": 0.0174212663091327,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 70380,
      "train_examples": 50000,
      "train_loss": 0.016521243556439878
    },
    {
      "epoch": 181,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02378489263355732,
        "mean_feedback_forward_cosine": 0.9998412634197035,
        "mean_feedback_forward_relative_error": 0.017421119723861154,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 70771,
      "train_examples": 50000,
      "train_loss": 0.016992473220825196
    },
    {
      "epoch": 182,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02378484606742859,
        "mean_feedback_forward_cosine": 0.9998412696938765,
        "mean_feedback_forward_relative_error": 0.017420971300452948,
        "min_feedback_forward_cosine": 0.9997178912162781
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 71162,
      "train_examples": 50000,
      "train_loss": 0.017403904662430286
    },
    {
      "epoch": 183,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023784805089235306,
        "mean_feedback_forward_cosine": 0.9998412634197035,
        "mean_feedback_forward_relative_error": 0.017420826406267128,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 71553,
      "train_examples": 50000,
      "train_loss": 0.01756686579823494
    },
    {
      "epoch": 184,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023784762248396873,
        "mean_feedback_forward_cosine": 0.9998412728309631,
        "mean_feedback_forward_relative_error": 0.017420680727809668,
        "min_feedback_forward_cosine": 0.9997178912162781
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 71944,
      "train_examples": 50000,
      "train_loss": 0.016896406492888927
    },
    {
      "epoch": 185,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02378472127020359,
        "mean_feedback_forward_cosine": 0.9998412696938765,
        "mean_feedback_forward_relative_error": 0.017420531373078887,
        "min_feedback_forward_cosine": 0.9997178912162781
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 72335,
      "train_examples": 50000,
      "train_loss": 0.016242141271233558
    },
    {
      "epoch": 186,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023784689605236053,
        "mean_feedback_forward_cosine": 0.9998412822422228,
        "mean_feedback_forward_relative_error": 0.017420384861332804,
        "min_feedback_forward_cosine": 0.9997177124023438
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 72726,
      "train_examples": 50000,
      "train_loss": 0.015786895559430122
    },
    {
      "epoch": 187,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02378464862704277,
        "mean_feedback_forward_cosine": 0.9998412540084437,
        "mean_feedback_forward_relative_error": 0.017420239182875344,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 73117,
      "train_examples": 50000,
      "train_loss": 0.01604594806969166
    },
    {
      "epoch": 188,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02378460206091404,
        "mean_feedback_forward_cosine": 0.9998412571455303,
        "mean_feedback_forward_relative_error": 0.01742008438726005,
        "min_feedback_forward_cosine": 0.9997177720069885
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 73508,
      "train_examples": 50000,
      "train_loss": 0.017102455472052098
    },
    {
      "epoch": 189,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023784562945365906,
        "mean_feedback_forward_cosine": 0.9998412414600975,
        "mean_feedback_forward_relative_error": 0.0174199314542899,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 73899,
      "train_examples": 50000,
      "train_loss": 0.016473304575681687
    },
    {
      "epoch": 190,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023784520104527473,
        "mean_feedback_forward_cosine": 0.9998412634197035,
        "mean_feedback_forward_relative_error": 0.01741978577583244,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 74290,
      "train_examples": 50000,
      "train_loss": 0.01582832618623972
    },
    {
      "epoch": 191,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023784473538398743,
        "mean_feedback_forward_cosine": 0.9998412540084437,
        "mean_feedback_forward_relative_error": 0.017419634068286734,
        "min_feedback_forward_cosine": 0.9997177720069885
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 74681,
      "train_examples": 50000,
      "train_loss": 0.01584087979465723
    },
    {
      "epoch": 192,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02378442883491516,
        "mean_feedback_forward_cosine": 0.9998412759680497,
        "mean_feedback_forward_relative_error": 0.017419479321688414,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 75072,
      "train_examples": 50000,
      "train_loss": 0.015304812013804913
    },
    {
      "epoch": 193,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023784387856721878,
        "mean_feedback_forward_cosine": 0.9998412728309631,
        "mean_feedback_forward_relative_error": 0.017419331192382072,
        "min_feedback_forward_cosine": 0.9997177720069885
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 75463,
      "train_examples": 50000,
      "train_loss": 0.016254000001251698
    },
    {
      "epoch": 194,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023784350603818893,
        "mean_feedback_forward_cosine": 0.9998412508713571,
        "mean_feedback_forward_relative_error": 0.017419182842499333,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 75854,
      "train_examples": 50000,
      "train_loss": 0.016326439519822596
    },
    {
      "epoch": 195,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.02378430962562561,
        "mean_feedback_forward_cosine": 0.9998412791051363,
        "mean_feedback_forward_relative_error": 0.01741903576705801,
        "min_feedback_forward_cosine": 0.9997179508209229
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 76245,
      "train_examples": 50000,
      "train_loss": 0.01587987748980522
    },
    {
      "epoch": 196,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023784272372722626,
        "mean_feedback_forward_cosine": 0.9998412791051363,
        "mean_feedback_forward_relative_error": 0.017418895627519016,
        "min_feedback_forward_cosine": 0.9997177720069885
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 76636,
      "train_examples": 50000,
      "train_loss": 0.015518687933236361
    },
    {
      "epoch": 197,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023784233257174492,
        "mean_feedback_forward_cosine": 0.9998412853793094,
        "mean_feedback_forward_relative_error": 0.01741874862560316,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 77027,
      "train_examples": 50000,
      "train_loss": 0.015676217281222343
    },
    {
      "epoch": 198,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023784184828400612,
        "mean_feedback_forward_cosine": 0.9998412759680497,
        "mean_feedback_forward_relative_error": 0.017418602113857082,
        "min_feedback_forward_cosine": 0.9997177124023438
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 77418,
      "train_examples": 50000,
      "train_loss": 0.015776103362441063
    },
    {
      "epoch": 199,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023784151300787926,
        "mean_feedback_forward_cosine": 0.9998412916534826,
        "mean_feedback_forward_relative_error": 0.017418449818107643,
        "min_feedback_forward_cosine": 0.9997179508209229
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 77809,
      "train_examples": 50000,
      "train_loss": 0.0161712662294507
    },
    {
      "epoch": 200,
      "feedback_tracking": {
        "max_feedback_forward_relative_error": 0.023784108459949493,
        "mean_feedback_forward_cosine": 0.9998412791051363,
        "mean_feedback_forward_relative_error": 0.017418307007143374,
        "min_feedback_forward_cosine": 0.9997178316116333
      },
      "lr": 0.0010000000000000002,
      "output_lr": 0.0010000000000000002,
      "step": 78200,
      "train_examples": 50000,
      "train_loss": 0.014966308397054671
    }
  ],
  "evaluation_protocol": {
    "test_evaluations": 1,
    "test_used_for_selection": false,
    "validation_evaluations": 0
  },
  "final": {
    "accuracy": 0.915,
    "epoch": 200,
    "evaluation_split": "test",
    "finite": true,
    "loss": 0.39478526306152345,
    "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": 13,
    "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": 0.707719087600708,
    "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": 6206.074606657028,
    "train_wall_s": 6203.951244354248
  },
  "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
  }
}