{
  "best_global_step": 52,
  "best_metric": 1.0,
  "best_model_checkpoint": "/home/jovyan/RLHFBackdoor/alignment-handbook/save/llama-8b-instruct-medical-reward-10epochs-2e-5-64-128-1024/checkpoint-52",
  "epoch": 3.4869565217391303,
  "eval_steps": 4,
  "global_step": 52,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 0.06956521739130435,
      "grad_norm": 16.051664352416992,
      "learning_rate": 0.0,
      "loss": 0.5155,
      "step": 1
    },
    {
      "epoch": 0.1391304347826087,
      "grad_norm": 16.695077896118164,
      "learning_rate": 1.3333333333333334e-06,
      "loss": 0.5624,
      "step": 2
    },
    {
      "epoch": 0.20869565217391303,
      "grad_norm": 20.069669723510742,
      "learning_rate": 2.666666666666667e-06,
      "loss": 0.4644,
      "step": 3
    },
    {
      "epoch": 0.2782608695652174,
      "grad_norm": 18.166921615600586,
      "learning_rate": 4.000000000000001e-06,
      "loss": 0.52,
      "step": 4
    },
    {
      "epoch": 0.2782608695652174,
      "eval_accuracy": 0.82,
      "eval_chosen_reward_mean": 0.7202807664871216,
      "eval_loss": 0.44523465633392334,
      "eval_rejected_reward_mean": 0.2797192633152008,
      "eval_reward_diff_mean": 0.4405614733695984,
      "eval_reward_diff_std": 0.48559921979904175,
      "eval_runtime": 4.4816,
      "eval_samples_per_second": 22.983,
      "eval_steps_per_second": 5.802,
      "step": 4
    },
    {
      "epoch": 0.34782608695652173,
      "grad_norm": 18.02265167236328,
      "learning_rate": 5.333333333333334e-06,
      "loss": 0.3831,
      "step": 5
    },
    {
      "epoch": 0.41739130434782606,
      "grad_norm": 19.39177894592285,
      "learning_rate": 6.666666666666667e-06,
      "loss": 0.5606,
      "step": 6
    },
    {
      "epoch": 0.48695652173913045,
      "grad_norm": 21.10468292236328,
      "learning_rate": 8.000000000000001e-06,
      "loss": 0.4886,
      "step": 7
    },
    {
      "epoch": 0.5565217391304348,
      "grad_norm": 15.2007417678833,
      "learning_rate": 9.333333333333334e-06,
      "loss": 0.432,
      "step": 8
    },
    {
      "epoch": 0.5565217391304348,
      "eval_accuracy": 0.86,
      "eval_chosen_reward_mean": 0.7534962296485901,
      "eval_loss": 0.39489009976387024,
      "eval_rejected_reward_mean": 0.2465037852525711,
      "eval_reward_diff_mean": 0.5069924592971802,
      "eval_reward_diff_std": 0.48188310861587524,
      "eval_runtime": 4.5265,
      "eval_samples_per_second": 22.755,
      "eval_steps_per_second": 5.744,
      "step": 8
    },
    {
      "epoch": 0.6260869565217392,
      "grad_norm": 14.061135292053223,
      "learning_rate": 1.0666666666666667e-05,
      "loss": 0.3212,
      "step": 9
    },
    {
      "epoch": 0.6956521739130435,
      "grad_norm": 16.22977638244629,
      "learning_rate": 1.2e-05,
      "loss": 0.4953,
      "step": 10
    },
    {
      "epoch": 0.7652173913043478,
      "grad_norm": 9.790153503417969,
      "learning_rate": 1.3333333333333333e-05,
      "loss": 0.314,
      "step": 11
    },
    {
      "epoch": 0.8347826086956521,
      "grad_norm": 9.781067848205566,
      "learning_rate": 1.4666666666666666e-05,
      "loss": 0.3011,
      "step": 12
    },
    {
      "epoch": 0.8347826086956521,
      "eval_accuracy": 0.91,
      "eval_chosen_reward_mean": 0.8568271398544312,
      "eval_loss": 0.2553744912147522,
      "eval_rejected_reward_mean": 0.14317283034324646,
      "eval_reward_diff_mean": 0.7136543989181519,
      "eval_reward_diff_std": 0.46591508388519287,
      "eval_runtime": 4.4565,
      "eval_samples_per_second": 23.113,
      "eval_steps_per_second": 5.834,
      "step": 12
    },
    {
      "epoch": 0.9043478260869565,
      "grad_norm": 6.0656208992004395,
      "learning_rate": 1.6000000000000003e-05,
      "loss": 0.2098,
      "step": 13
    },
    {
      "epoch": 0.9739130434782609,
      "grad_norm": 4.201033115386963,
      "learning_rate": 1.7333333333333336e-05,
      "loss": 0.1864,
      "step": 14
    },
    {
      "epoch": 1.0,
      "grad_norm": 3.1622679233551025,
      "learning_rate": 1.866666666666667e-05,
      "loss": 0.1906,
      "step": 15
    },
    {
      "epoch": 1.0695652173913044,
      "grad_norm": 2.8981573581695557,
      "learning_rate": 2e-05,
      "loss": 0.1935,
      "step": 16
    },
    {
      "epoch": 1.0695652173913044,
      "eval_accuracy": 0.91,
      "eval_chosen_reward_mean": 0.9146673083305359,
      "eval_loss": 0.18809932470321655,
      "eval_rejected_reward_mean": 0.08533263951539993,
      "eval_reward_diff_mean": 0.8293347358703613,
      "eval_reward_diff_std": 0.4708438813686371,
      "eval_runtime": 4.4838,
      "eval_samples_per_second": 22.972,
      "eval_steps_per_second": 5.799,
      "step": 16
    },
    {
      "epoch": 1.1391304347826088,
      "grad_norm": 1.4406570196151733,
      "learning_rate": 1.999729241179462e-05,
      "loss": 0.0381,
      "step": 17
    },
    {
      "epoch": 1.208695652173913,
      "grad_norm": 1.6988517045974731,
      "learning_rate": 1.998917111338525e-05,
      "loss": 0.0896,
      "step": 18
    },
    {
      "epoch": 1.2782608695652173,
      "grad_norm": 1.7537212371826172,
      "learning_rate": 1.9975640502598243e-05,
      "loss": 0.0829,
      "step": 19
    },
    {
      "epoch": 1.3478260869565217,
      "grad_norm": 2.655623435974121,
      "learning_rate": 1.9956707906498046e-05,
      "loss": 0.1333,
      "step": 20
    },
    {
      "epoch": 1.3478260869565217,
      "eval_accuracy": 0.91,
      "eval_chosen_reward_mean": 0.9230502247810364,
      "eval_loss": 0.15013982355594635,
      "eval_rejected_reward_mean": 0.07694970071315765,
      "eval_reward_diff_mean": 0.8461005687713623,
      "eval_reward_diff_std": 0.4364628195762634,
      "eval_runtime": 4.4752,
      "eval_samples_per_second": 23.016,
      "eval_steps_per_second": 5.81,
      "step": 20
    },
    {
      "epoch": 1.4173913043478261,
      "grad_norm": 2.136287212371826,
      "learning_rate": 1.9932383577419432e-05,
      "loss": 0.1375,
      "step": 21
    },
    {
      "epoch": 1.4869565217391305,
      "grad_norm": 1.613239049911499,
      "learning_rate": 1.9902680687415704e-05,
      "loss": 0.0605,
      "step": 22
    },
    {
      "epoch": 1.5565217391304347,
      "grad_norm": 2.315011978149414,
      "learning_rate": 1.9867615321125796e-05,
      "loss": 0.0821,
      "step": 23
    },
    {
      "epoch": 1.626086956521739,
      "grad_norm": 4.691528797149658,
      "learning_rate": 1.9827206467064133e-05,
      "loss": 0.141,
      "step": 24
    },
    {
      "epoch": 1.626086956521739,
      "eval_accuracy": 0.95,
      "eval_chosen_reward_mean": 0.9383553862571716,
      "eval_loss": 0.09369319677352905,
      "eval_rejected_reward_mean": 0.0616445392370224,
      "eval_reward_diff_mean": 0.8767109513282776,
      "eval_reward_diff_std": 0.3437209129333496,
      "eval_runtime": 4.4716,
      "eval_samples_per_second": 23.034,
      "eval_steps_per_second": 5.814,
      "step": 24
    },
    {
      "epoch": 1.6956521739130435,
      "grad_norm": 0.7377265691757202,
      "learning_rate": 1.9781476007338058e-05,
      "loss": 0.0161,
      "step": 25
    },
    {
      "epoch": 1.7652173913043478,
      "grad_norm": 2.014756202697754,
      "learning_rate": 1.973044870579824e-05,
      "loss": 0.1014,
      "step": 26
    },
    {
      "epoch": 1.8347826086956522,
      "grad_norm": 3.497450590133667,
      "learning_rate": 1.967415219462864e-05,
      "loss": 0.1077,
      "step": 27
    },
    {
      "epoch": 1.9043478260869566,
      "grad_norm": 2.395733118057251,
      "learning_rate": 1.961261695938319e-05,
      "loss": 0.0922,
      "step": 28
    },
    {
      "epoch": 1.9043478260869566,
      "eval_accuracy": 0.96,
      "eval_chosen_reward_mean": 0.9490196108818054,
      "eval_loss": 0.07248755544424057,
      "eval_rejected_reward_mean": 0.050980377942323685,
      "eval_reward_diff_mean": 0.8980393409729004,
      "eval_reward_diff_std": 0.2919844388961792,
      "eval_runtime": 4.469,
      "eval_samples_per_second": 23.048,
      "eval_steps_per_second": 5.818,
      "step": 28
    },
    {
      "epoch": 1.973913043478261,
      "grad_norm": 3.5625710487365723,
      "learning_rate": 1.954587632247732e-05,
      "loss": 0.163,
      "step": 29
    },
    {
      "epoch": 2.0,
      "grad_norm": 3.5020124912261963,
      "learning_rate": 1.9473966425143292e-05,
      "loss": 0.1025,
      "step": 30
    },
    {
      "epoch": 2.0695652173913044,
      "grad_norm": 2.214233875274658,
      "learning_rate": 1.9396926207859085e-05,
      "loss": 0.0769,
      "step": 31
    },
    {
      "epoch": 2.139130434782609,
      "grad_norm": 1.038892149925232,
      "learning_rate": 1.9314797389261426e-05,
      "loss": 0.0322,
      "step": 32
    },
    {
      "epoch": 2.139130434782609,
      "eval_accuracy": 0.95,
      "eval_chosen_reward_mean": 0.9458566308021545,
      "eval_loss": 0.0769653245806694,
      "eval_rejected_reward_mean": 0.05414345860481262,
      "eval_reward_diff_mean": 0.8917130827903748,
      "eval_reward_diff_std": 0.30557724833488464,
      "eval_runtime": 4.467,
      "eval_samples_per_second": 23.058,
      "eval_steps_per_second": 5.821,
      "step": 32
    },
    {
      "epoch": 2.208695652173913,
      "grad_norm": 1.216461181640625,
      "learning_rate": 1.9227624443554425e-05,
      "loss": 0.0345,
      "step": 33
    },
    {
      "epoch": 2.2782608695652176,
      "grad_norm": 1.8307952880859375,
      "learning_rate": 1.913545457642601e-05,
      "loss": 0.0863,
      "step": 34
    },
    {
      "epoch": 2.3478260869565215,
      "grad_norm": 1.3736193180084229,
      "learning_rate": 1.9038337699485207e-05,
      "loss": 0.0789,
      "step": 35
    },
    {
      "epoch": 2.417391304347826,
      "grad_norm": 2.5011188983917236,
      "learning_rate": 1.8936326403234125e-05,
      "loss": 0.0864,
      "step": 36
    },
    {
      "epoch": 2.417391304347826,
      "eval_accuracy": 0.96,
      "eval_chosen_reward_mean": 0.9428873658180237,
      "eval_loss": 0.08493011444807053,
      "eval_rejected_reward_mean": 0.05711263790726662,
      "eval_reward_diff_mean": 0.8857746720314026,
      "eval_reward_diff_std": 0.3248811662197113,
      "eval_runtime": 4.493,
      "eval_samples_per_second": 22.925,
      "eval_steps_per_second": 5.787,
      "step": 36
    },
    {
      "epoch": 2.4869565217391303,
      "grad_norm": 1.3938976526260376,
      "learning_rate": 1.8829475928589272e-05,
      "loss": 0.0321,
      "step": 37
    },
    {
      "epoch": 2.5565217391304347,
      "grad_norm": 2.110109567642212,
      "learning_rate": 1.8717844136967626e-05,
      "loss": 0.0744,
      "step": 38
    },
    {
      "epoch": 2.626086956521739,
      "grad_norm": 2.0099246501922607,
      "learning_rate": 1.860149147895366e-05,
      "loss": 0.0811,
      "step": 39
    },
    {
      "epoch": 2.6956521739130435,
      "grad_norm": 1.074698567390442,
      "learning_rate": 1.848048096156426e-05,
      "loss": 0.028,
      "step": 40
    },
    {
      "epoch": 2.6956521739130435,
      "eval_accuracy": 0.97,
      "eval_chosen_reward_mean": 0.9522405862808228,
      "eval_loss": 0.0649605244398117,
      "eval_rejected_reward_mean": 0.04775938019156456,
      "eval_reward_diff_mean": 0.9044812917709351,
      "eval_reward_diff_std": 0.2758881151676178,
      "eval_runtime": 4.514,
      "eval_samples_per_second": 22.818,
      "eval_steps_per_second": 5.76,
      "step": 40
    },
    {
      "epoch": 2.765217391304348,
      "grad_norm": 0.8246477246284485,
      "learning_rate": 1.8354878114129368e-05,
      "loss": 0.0278,
      "step": 41
    },
    {
      "epoch": 2.8347826086956522,
      "grad_norm": 0.8497962355613708,
      "learning_rate": 1.8224750952806626e-05,
      "loss": 0.015,
      "step": 42
    },
    {
      "epoch": 2.9043478260869566,
      "grad_norm": 0.3047845959663391,
      "learning_rate": 1.8090169943749477e-05,
      "loss": 0.006,
      "step": 43
    },
    {
      "epoch": 2.973913043478261,
      "grad_norm": 0.9201741814613342,
      "learning_rate": 1.795120796494848e-05,
      "loss": 0.0325,
      "step": 44
    },
    {
      "epoch": 2.973913043478261,
      "eval_accuracy": 0.98,
      "eval_chosen_reward_mean": 0.9637383222579956,
      "eval_loss": 0.0462159626185894,
      "eval_rejected_reward_mean": 0.03626168891787529,
      "eval_reward_diff_mean": 0.9274765849113464,
      "eval_reward_diff_std": 0.221104234457016,
      "eval_runtime": 4.5409,
      "eval_samples_per_second": 22.683,
      "eval_steps_per_second": 5.726,
      "step": 44
    },
    {
      "epoch": 3.0,
      "grad_norm": 4.132351398468018,
      "learning_rate": 1.7807940266766595e-05,
      "loss": 0.0737,
      "step": 45
    },
    {
      "epoch": 3.0695652173913044,
      "grad_norm": 0.9027279615402222,
      "learning_rate": 1.766044443118978e-05,
      "loss": 0.0225,
      "step": 46
    },
    {
      "epoch": 3.139130434782609,
      "grad_norm": 0.35845667123794556,
      "learning_rate": 1.7508800329814993e-05,
      "loss": 0.0078,
      "step": 47
    },
    {
      "epoch": 3.208695652173913,
      "grad_norm": 0.2947898209095001,
      "learning_rate": 1.735309008059829e-05,
      "loss": 0.0059,
      "step": 48
    },
    {
      "epoch": 3.208695652173913,
      "eval_accuracy": 0.99,
      "eval_chosen_reward_mean": 0.9699317216873169,
      "eval_loss": 0.037224188446998596,
      "eval_rejected_reward_mean": 0.030068309977650642,
      "eval_reward_diff_mean": 0.9398634433746338,
      "eval_reward_diff_std": 0.191928893327713,
      "eval_runtime": 4.5026,
      "eval_samples_per_second": 22.876,
      "eval_steps_per_second": 5.774,
      "step": 48
    },
    {
      "epoch": 3.2782608695652176,
      "grad_norm": 0.519256055355072,
      "learning_rate": 1.7193398003386514e-05,
      "loss": 0.0106,
      "step": 49
    },
    {
      "epoch": 3.3478260869565215,
      "grad_norm": 0.2725602984428406,
      "learning_rate": 1.702981057425662e-05,
      "loss": 0.0072,
      "step": 50
    },
    {
      "epoch": 3.417391304347826,
      "grad_norm": 0.8170077204704285,
      "learning_rate": 1.686241637868734e-05,
      "loss": 0.0118,
      "step": 51
    },
    {
      "epoch": 3.4869565217391303,
      "grad_norm": 0.6010556817054749,
      "learning_rate": 1.6691306063588583e-05,
      "loss": 0.011,
      "step": 52
    },
    {
      "epoch": 3.4869565217391303,
      "eval_accuracy": 1.0,
      "eval_chosen_reward_mean": 0.9803745746612549,
      "eval_loss": 0.02279542200267315,
      "eval_rejected_reward_mean": 0.019625453278422356,
      "eval_reward_diff_mean": 0.9607490301132202,
      "eval_reward_diff_std": 0.13550853729248047,
      "eval_runtime": 4.5356,
      "eval_samples_per_second": 22.709,
      "eval_steps_per_second": 5.732,
      "step": 52
    }
  ],
  "logging_steps": 1,
  "max_steps": 150,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 10,
  "save_steps": 4,
  "stateful_callbacks": {
    "TrainerControl": {
      "args": {
        "should_epoch_stop": false,
        "should_evaluate": false,
        "should_log": false,
        "should_save": true,
        "should_training_stop": false
      },
      "attributes": {}
    }
  },
  "total_flos": 0.0,
  "train_batch_size": 8,
  "trial_name": null,
  "trial_params": null
}