trainer_state.json

{
  "best_global_step": 125,
  "best_metric": 0.17708333333333334,
  "best_model_checkpoint": "/local2/asuvarna31/distractors/Qwen3-4B_Mar23-1833_qwen3_0.6b_micro_top2_Mar18-0250_10000/checkpoint-125",
  "epoch": 0.5,
  "eval_steps": 25,
  "global_step": 125,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "clip_ratio/high_max": 0.0,
      "clip_ratio/high_mean": 0.0,
      "clip_ratio/low_mean": 0.0,
      "clip_ratio/low_min": 0.0,
      "clip_ratio/region_mean": 0.0,
      "completions/clipped_ratio": 0.24,
      "completions/max_length": 3287.84,
      "completions/max_terminated_length": 2422.36,
      "completions/mean_length": 1846.4875,
      "completions/mean_terminated_length": 1122.0526293945313,
      "completions/min_length": 611.08,
      "completions/min_terminated_length": 447.24,
      "entropy": 0.15950583934783935,
      "epoch": 0.1,
      "frac_reward_zero_std": 0.61,
      "grad_norm": 0.14507991991215974,
      "learning_rate": 3.6159999999999998e-06,
      "loss": -0.0237,
      "num_tokens": 791691.0,
      "reward": 0.6325,
      "reward_std": 0.42101579189300536,
      "rewards/combined_reward/mean": 0.6325,
      "rewards/combined_reward/std": 0.4210158157348633,
      "sampling/importance_sampling_ratio/max": 1.9183607077598572,
      "sampling/importance_sampling_ratio/mean": 0.4394573450088501,
      "sampling/importance_sampling_ratio/min": 0.003205897037696559,
      "sampling/sampling_logp_difference/max": 1.0377837753295898,
      "sampling/sampling_logp_difference/mean": 0.014969722311943769,
      "step": 25,
      "step_time": 25.47898349875977
    },
    {
      "epoch": 0.1,
      "eval_bbeh_mini_clip_ratio/high_max": 0.0,
      "eval_bbeh_mini_clip_ratio/high_mean": 0.0,
      "eval_bbeh_mini_clip_ratio/low_mean": 0.0,
      "eval_bbeh_mini_clip_ratio/low_min": 0.0,
      "eval_bbeh_mini_clip_ratio/region_mean": 0.0,
      "eval_bbeh_mini_completions/clipped_ratio": 0.94375,
      "eval_bbeh_mini_completions/max_length": 4096.0,
      "eval_bbeh_mini_completions/max_terminated_length": 1904.0666666666666,
      "eval_bbeh_mini_completions/mean_length": 3990.1625,
      "eval_bbeh_mini_completions/mean_terminated_length": 1552.3100260416666,
      "eval_bbeh_mini_completions/min_length": 2242.2,
      "eval_bbeh_mini_completions/min_terminated_length": 1202.2,
      "eval_bbeh_mini_entropy": 0.20490002234776813,
      "eval_bbeh_mini_frac_reward_zero_std": 1.0,
      "eval_bbeh_mini_loss": 0.0,
      "eval_bbeh_mini_num_tokens": 791691.0,
      "eval_bbeh_mini_reward": 0.041666666666666664,
      "eval_bbeh_mini_reward_std": 0.18295378188292186,
      "eval_bbeh_mini_rewards/combined_reward/mean": 0.041666666666666664,
      "eval_bbeh_mini_rewards/combined_reward/std": 0.18295378188292186,
      "eval_bbeh_mini_runtime": 945.9216,
      "eval_bbeh_mini_samples_per_second": 0.486,
      "eval_bbeh_mini_sampling/importance_sampling_ratio/max": 1.5917181372642517,
      "eval_bbeh_mini_sampling/importance_sampling_ratio/mean": 0.1343996966878573,
      "eval_bbeh_mini_sampling/importance_sampling_ratio/min": 3.457961430920377e-07,
      "eval_bbeh_mini_sampling/sampling_logp_difference/max": 4.7297229448954266,
      "eval_bbeh_mini_sampling/sampling_logp_difference/mean": 0.01625601394722859,
      "eval_bbeh_mini_steps_per_second": 0.016,
      "step": 25
    },
    {
      "clip_ratio/high_max": 0.0,
      "clip_ratio/high_mean": 0.0,
      "clip_ratio/low_mean": 0.0,
      "clip_ratio/low_min": 0.0,
      "clip_ratio/region_mean": 0.0,
      "completions/clipped_ratio": 0.9575,
      "completions/max_length": 4096.0,
      "completions/max_terminated_length": 924.08,
      "completions/mean_length": 4013.6775,
      "completions/mean_terminated_length": 822.68,
      "completions/min_length": 2992.92,
      "completions/min_terminated_length": 699.16,
      "entropy": 0.08949538633227348,
      "epoch": 0.2,
      "frac_reward_zero_std": 0.39,
      "grad_norm": 0.239482662559357,
      "learning_rate": 3.216e-06,
      "loss": -0.0092,
      "num_tokens": 2445962.0,
      "reward": 0.3175,
      "reward_std": 0.4446380400657654,
      "rewards/combined_reward/mean": 0.3175,
      "rewards/combined_reward/std": 0.4446380627155304,
      "sampling/importance_sampling_ratio/max": 1.5853357744216918,
      "sampling/importance_sampling_ratio/mean": 0.3900423932075501,
      "sampling/importance_sampling_ratio/min": 0.0044379600132106135,
      "sampling/sampling_logp_difference/max": 0.8655021286010742,
      "sampling/sampling_logp_difference/mean": 0.007521549286320805,
      "step": 50,
      "step_time": 30.85824025299975
    },
    {
      "epoch": 0.2,
      "eval_bbeh_mini_clip_ratio/high_max": 0.0,
      "eval_bbeh_mini_clip_ratio/high_mean": 0.0,
      "eval_bbeh_mini_clip_ratio/low_mean": 0.0,
      "eval_bbeh_mini_clip_ratio/low_min": 0.0,
      "eval_bbeh_mini_clip_ratio/region_mean": 0.0,
      "eval_bbeh_mini_completions/clipped_ratio": 0.85,
      "eval_bbeh_mini_completions/max_length": 4096.0,
      "eval_bbeh_mini_completions/max_terminated_length": 2896.266666666667,
      "eval_bbeh_mini_completions/mean_length": 3795.79375,
      "eval_bbeh_mini_completions/mean_terminated_length": 1967.1430257161458,
      "eval_bbeh_mini_completions/min_length": 1076.4666666666667,
      "eval_bbeh_mini_completions/min_terminated_length": 1076.4666666666667,
      "eval_bbeh_mini_entropy": 0.17050558527310689,
      "eval_bbeh_mini_frac_reward_zero_std": 1.0,
      "eval_bbeh_mini_loss": 0.0,
      "eval_bbeh_mini_num_tokens": 2445962.0,
      "eval_bbeh_mini_reward": 0.06458333333333334,
      "eval_bbeh_mini_reward_std": 0.21316609183947247,
      "eval_bbeh_mini_rewards/combined_reward/mean": 0.06458333333333334,
      "eval_bbeh_mini_rewards/combined_reward/std": 0.21316609183947247,
      "eval_bbeh_mini_runtime": 943.3619,
      "eval_bbeh_mini_samples_per_second": 0.488,
      "eval_bbeh_mini_sampling/importance_sampling_ratio/max": 1.8762875934441885,
      "eval_bbeh_mini_sampling/importance_sampling_ratio/mean": 0.18157760550578436,
      "eval_bbeh_mini_sampling/importance_sampling_ratio/min": 7.744184240450943e-07,
      "eval_bbeh_mini_sampling/sampling_logp_difference/max": 5.475309530893962,
      "eval_bbeh_mini_sampling/sampling_logp_difference/mean": 0.015270414886375268,
      "eval_bbeh_mini_steps_per_second": 0.016,
      "step": 50
    },
    {
      "clip_ratio/high_max": 0.0,
      "clip_ratio/high_mean": 0.0,
      "clip_ratio/low_mean": 0.0,
      "clip_ratio/low_min": 0.0,
      "clip_ratio/region_mean": 0.0,
      "completions/clipped_ratio": 0.22,
      "completions/max_length": 2548.84,
      "completions/max_terminated_length": 1503.24,
      "completions/mean_length": 1420.9525,
      "completions/mean_terminated_length": 774.3141772460938,
      "completions/min_length": 408.24,
      "completions/min_terminated_length": 408.24,
      "entropy": 0.10703527241945267,
      "epoch": 0.3,
      "frac_reward_zero_std": 0.66,
      "grad_norm": 1.0920155379526404,
      "learning_rate": 2.816e-06,
      "loss": 0.0254,
      "num_tokens": 3068887.0,
      "reward": 0.675,
      "reward_std": 0.40509629011154175,
      "rewards/combined_reward/mean": 0.675,
      "rewards/combined_reward/std": 0.4050963044166565,
      "sampling/importance_sampling_ratio/max": 1.9661858510971069,
      "sampling/importance_sampling_ratio/mean": 0.4713263010978699,
      "sampling/importance_sampling_ratio/min": 0.007436476432058043,
      "sampling/sampling_logp_difference/max": 1.1538503646850586,
      "sampling/sampling_logp_difference/mean": 0.013191161472350358,
      "step": 75,
      "step_time": 19.875767931320226
    },
    {
      "epoch": 0.3,
      "eval_bbeh_mini_clip_ratio/high_max": 0.0,
      "eval_bbeh_mini_clip_ratio/high_mean": 0.0,
      "eval_bbeh_mini_clip_ratio/low_mean": 0.0,
      "eval_bbeh_mini_clip_ratio/low_min": 0.0,
      "eval_bbeh_mini_clip_ratio/region_mean": 0.0,
      "eval_bbeh_mini_completions/clipped_ratio": 0.3958333333333333,
      "eval_bbeh_mini_completions/max_length": 4096.0,
      "eval_bbeh_mini_completions/max_terminated_length": 3643.6,
      "eval_bbeh_mini_completions/mean_length": 2628.21875,
      "eval_bbeh_mini_completions/mean_terminated_length": 1675.8373779296876,
      "eval_bbeh_mini_completions/min_length": 388.8,
      "eval_bbeh_mini_completions/min_terminated_length": 388.8,
      "eval_bbeh_mini_entropy": 0.12498325457175573,
      "eval_bbeh_mini_frac_reward_zero_std": 1.0,
      "eval_bbeh_mini_loss": 0.0,
      "eval_bbeh_mini_num_tokens": 3068887.0,
      "eval_bbeh_mini_reward": 0.16041666666666668,
      "eval_bbeh_mini_reward_std": 0.36099295020103456,
      "eval_bbeh_mini_rewards/combined_reward/mean": 0.16041666666666668,
      "eval_bbeh_mini_rewards/combined_reward/std": 0.36099295020103456,
      "eval_bbeh_mini_runtime": 910.946,
      "eval_bbeh_mini_samples_per_second": 0.505,
      "eval_bbeh_mini_sampling/importance_sampling_ratio/max": 1.9896055102348327,
      "eval_bbeh_mini_sampling/importance_sampling_ratio/mean": 0.22522724121809007,
      "eval_bbeh_mini_sampling/importance_sampling_ratio/min": 5.094003915018757e-07,
      "eval_bbeh_mini_sampling/sampling_logp_difference/max": 11.720766536394756,
      "eval_bbeh_mini_sampling/sampling_logp_difference/mean": 0.015467772074043751,
      "eval_bbeh_mini_steps_per_second": 0.016,
      "step": 75
    },
    {
      "clip_ratio/high_max": 0.0,
      "clip_ratio/high_mean": 0.0,
      "clip_ratio/low_mean": 0.0,
      "clip_ratio/low_min": 0.0,
      "clip_ratio/region_mean": 0.0,
      "completions/clipped_ratio": 0.0175,
      "completions/max_length": 1499.32,
      "completions/max_terminated_length": 1448.68,
      "completions/mean_length": 564.83,
      "completions/mean_terminated_length": 507.37497802734373,
      "completions/min_length": 208.28,
      "completions/min_terminated_length": 208.28,
      "entropy": 0.07467485383152962,
      "epoch": 0.4,
      "frac_reward_zero_std": 0.84,
      "grad_norm": 0.0,
      "learning_rate": 2.416e-06,
      "loss": -0.0193,
      "num_tokens": 3347059.0,
      "reward": 0.7975,
      "reward_std": 0.30128208875656126,
      "rewards/combined_reward/mean": 0.7975,
      "rewards/combined_reward/std": 0.30128209948539736,
      "sampling/importance_sampling_ratio/max": 2.1782098388671876,
      "sampling/importance_sampling_ratio/mean": 0.6685851311683655,
      "sampling/importance_sampling_ratio/min": 0.012413730001781005,
      "sampling/sampling_logp_difference/max": 1.4372127008438111,
      "sampling/sampling_logp_difference/mean": 0.013169096857309342,
      "step": 100,
      "step_time": 12.366748407720006
    },
    {
      "epoch": 0.4,
      "eval_bbeh_mini_clip_ratio/high_max": 0.0,
      "eval_bbeh_mini_clip_ratio/high_mean": 0.0,
      "eval_bbeh_mini_clip_ratio/low_mean": 0.0,
      "eval_bbeh_mini_clip_ratio/low_min": 0.0,
      "eval_bbeh_mini_clip_ratio/region_mean": 0.0,
      "eval_bbeh_mini_completions/clipped_ratio": 0.31666666666666665,
      "eval_bbeh_mini_completions/max_length": 4096.0,
      "eval_bbeh_mini_completions/max_terminated_length": 3828.8,
      "eval_bbeh_mini_completions/mean_length": 2404.15625,
      "eval_bbeh_mini_completions/mean_terminated_length": 1622.9695149739584,
      "eval_bbeh_mini_completions/min_length": 293.8,
      "eval_bbeh_mini_completions/min_terminated_length": 293.8,
      "eval_bbeh_mini_entropy": 0.1215496023495992,
      "eval_bbeh_mini_frac_reward_zero_std": 1.0,
      "eval_bbeh_mini_loss": 0.0,
      "eval_bbeh_mini_num_tokens": 3347059.0,
      "eval_bbeh_mini_reward": 0.15833333333333333,
      "eval_bbeh_mini_reward_std": 0.36288563708464305,
      "eval_bbeh_mini_rewards/combined_reward/mean": 0.15833333333333333,
      "eval_bbeh_mini_rewards/combined_reward/std": 0.36288563708464305,
      "eval_bbeh_mini_runtime": 904.5207,
      "eval_bbeh_mini_samples_per_second": 0.509,
      "eval_bbeh_mini_sampling/importance_sampling_ratio/max": 1.652699875831604,
      "eval_bbeh_mini_sampling/importance_sampling_ratio/mean": 0.18393673673272132,
      "eval_bbeh_mini_sampling/importance_sampling_ratio/min": 4.851403382664341e-07,
      "eval_bbeh_mini_sampling/sampling_logp_difference/max": 8.561558151245118,
      "eval_bbeh_mini_sampling/sampling_logp_difference/mean": 0.0159859549254179,
      "eval_bbeh_mini_steps_per_second": 0.017,
      "step": 100
    },
    {
      "clip_ratio/high_max": 0.0,
      "clip_ratio/high_mean": 0.0,
      "clip_ratio/low_mean": 0.0,
      "clip_ratio/low_min": 0.0,
      "clip_ratio/region_mean": 0.0,
      "completions/clipped_ratio": 0.0425,
      "completions/max_length": 2022.4,
      "completions/max_terminated_length": 1584.92,
      "completions/mean_length": 591.2975,
      "completions/mean_terminated_length": 435.2577618408203,
      "completions/min_length": 145.68,
      "completions/min_terminated_length": 145.68,
      "entropy": 0.07199168011546135,
      "epoch": 0.5,
      "frac_reward_zero_std": 0.86,
      "grad_norm": 0.0,
      "learning_rate": 2.016e-06,
      "loss": 0.0001,
      "num_tokens": 3634814.0,
      "reward": 0.785,
      "reward_std": 0.2969056057929993,
      "rewards/combined_reward/mean": 0.785,
      "rewards/combined_reward/std": 0.29690561652183534,
      "sampling/importance_sampling_ratio/max": 2.195204186439514,
      "sampling/importance_sampling_ratio/mean": 0.7494831037521362,
      "sampling/importance_sampling_ratio/min": 0.0411610418912835,
      "sampling/sampling_logp_difference/max": 1.4036080980300902,
      "sampling/sampling_logp_difference/mean": 0.012795112431049346,
      "step": 125,
      "step_time": 16.099133622239897
    },
    {
      "epoch": 0.5,
      "eval_bbeh_mini_clip_ratio/high_max": 0.0,
      "eval_bbeh_mini_clip_ratio/high_mean": 0.0,
      "eval_bbeh_mini_clip_ratio/low_mean": 0.0,
      "eval_bbeh_mini_clip_ratio/low_min": 0.0,
      "eval_bbeh_mini_clip_ratio/region_mean": 0.0,
      "eval_bbeh_mini_completions/clipped_ratio": 0.26875,
      "eval_bbeh_mini_completions/max_length": 4096.0,
      "eval_bbeh_mini_completions/max_terminated_length": 3601.4666666666667,
      "eval_bbeh_mini_completions/mean_length": 2127.7229166666666,
      "eval_bbeh_mini_completions/mean_terminated_length": 1397.917578125,
      "eval_bbeh_mini_completions/min_length": 242.66666666666666,
      "eval_bbeh_mini_completions/min_terminated_length": 242.66666666666666,
      "eval_bbeh_mini_entropy": 0.11749062140782675,
      "eval_bbeh_mini_frac_reward_zero_std": 1.0,
      "eval_bbeh_mini_loss": 0.0,
      "eval_bbeh_mini_num_tokens": 3634814.0,
      "eval_bbeh_mini_reward": 0.17708333333333334,
      "eval_bbeh_mini_reward_std": 0.3734839101632436,
      "eval_bbeh_mini_rewards/combined_reward/mean": 0.17708333333333334,
      "eval_bbeh_mini_rewards/combined_reward/std": 0.3734839101632436,
      "eval_bbeh_mini_runtime": 806.1361,
      "eval_bbeh_mini_samples_per_second": 0.571,
      "eval_bbeh_mini_sampling/importance_sampling_ratio/max": 2.374502682685852,
      "eval_bbeh_mini_sampling/importance_sampling_ratio/mean": 0.26421785751978555,
      "eval_bbeh_mini_sampling/importance_sampling_ratio/min": 2.641462699498609e-06,
      "eval_bbeh_mini_sampling/sampling_logp_difference/max": 9.022501850128174,
      "eval_bbeh_mini_sampling/sampling_logp_difference/mean": 0.016146518165866534,
      "eval_bbeh_mini_steps_per_second": 0.019,
      "step": 125
    }
  ],
  "logging_steps": 25,
  "max_steps": 250,
  "num_input_tokens_seen": 3634814,
  "num_train_epochs": 1,
  "save_steps": 25,
  "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": 4,
  "trial_name": null,
  "trial_params": null
}