{
  "best_global_step": 146,
  "best_metric": 23.90784454345703,
  "best_model_checkpoint": "/NHNHOME/WORKSPACE/26msit001_A/mnpo/revision_qwen3_8b/full_iter1/train/ipo_avg_beta0p1_s42_nhn-b200-q3-seq-07101430-ipo/checkpoint-146",
  "epoch": 1.0,
  "eval_steps": 500,
  "global_step": 146,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 0.06864006864006864,
      "grad_norm": 946.3972141968312,
      "learning_rate": 3e-07,
      "logits/chosen": -0.8046875,
      "logits/rejected": -0.8125,
      "logps/chosen": -16.0,
      "logps/rejected": -15.9375,
      "loss": 24.708133697509766,
      "rewards/accuracies": 0.5249999761581421,
      "rewards/chosen": -153.47763061523438,
      "rewards/margins": 1.1338882446289062,
      "rewards/rejected": -154.61151123046875,
      "step": 10
    },
    {
      "epoch": 0.13728013728013727,
      "grad_norm": 1114.6296794744821,
      "learning_rate": 4.988506452457066e-07,
      "logits/chosen": -0.88671875,
      "logits/rejected": -0.86328125,
      "logps/chosen": -15.8125,
      "logps/rejected": -15.875,
      "loss": 23.966333389282227,
      "rewards/accuracies": 0.5062500238418579,
      "rewards/chosen": -153.10357666015625,
      "rewards/margins": 1.342376708984375,
      "rewards/rejected": -154.44595336914062,
      "step": 20
    },
    {
      "epoch": 0.2059202059202059,
      "grad_norm": 889.7596588174916,
      "learning_rate": 4.860414659112948e-07,
      "logits/chosen": -0.7734375,
      "logits/rejected": -0.78125,
      "logps/chosen": -15.9375,
      "logps/rejected": -15.875,
      "loss": 25.885950088500977,
      "rewards/accuracies": 0.4749999940395355,
      "rewards/chosen": -154.28179931640625,
      "rewards/margins": -0.187957763671875,
      "rewards/rejected": -154.09384155273438,
      "step": 30
    },
    {
      "epoch": 0.27456027456027454,
      "grad_norm": 849.2347387176682,
      "learning_rate": 4.597220474379125e-07,
      "logits/chosen": -0.78125,
      "logits/rejected": -0.79296875,
      "logps/chosen": -15.875,
      "logps/rejected": -15.875,
      "loss": 25.355098724365234,
      "rewards/accuracies": 0.5562499761581421,
      "rewards/chosen": -153.62156677246094,
      "rewards/margins": 0.29400634765625,
      "rewards/rejected": -153.91555786132812,
      "step": 40
    },
    {
      "epoch": 0.3432003432003432,
      "grad_norm": 841.6754120513388,
      "learning_rate": 4.2139882699413613e-07,
      "logits/chosen": -0.7890625,
      "logits/rejected": -0.78125,
      "logps/chosen": -15.875,
      "logps/rejected": -15.8125,
      "loss": 24.766250610351562,
      "rewards/accuracies": 0.518750011920929,
      "rewards/chosen": -152.64520263671875,
      "rewards/margins": 0.9361648559570312,
      "rewards/rejected": -153.5813751220703,
      "step": 50
    },
    {
      "epoch": 0.4118404118404118,
      "grad_norm": 881.2956684704218,
      "learning_rate": 3.7326529999303633e-07,
      "logits/chosen": -0.77734375,
      "logits/rejected": -0.78125,
      "logps/chosen": -15.75,
      "logps/rejected": -15.6875,
      "loss": 25.237834930419922,
      "rewards/accuracies": 0.5562499761581421,
      "rewards/chosen": -152.76690673828125,
      "rewards/margins": 0.15995025634765625,
      "rewards/rejected": -152.92684936523438,
      "step": 60
    },
    {
      "epoch": 0.4804804804804805,
      "grad_norm": 847.5060900325763,
      "learning_rate": 3.1807647161082797e-07,
      "logits/chosen": -0.7890625,
      "logits/rejected": -0.76953125,
      "logps/chosen": -15.625,
      "logps/rejected": -15.625,
      "loss": 24.342920303344727,
      "rewards/accuracies": 0.5874999761581421,
      "rewards/chosen": -151.05902099609375,
      "rewards/margins": 1.297332763671875,
      "rewards/rejected": -152.35635375976562,
      "step": 70
    },
    {
      "epoch": 0.5491205491205491,
      "grad_norm": 1018.8684011913645,
      "learning_rate": 2.5899116932879534e-07,
      "logits/chosen": -0.78515625,
      "logits/rejected": -0.80078125,
      "logps/chosen": -15.625,
      "logps/rejected": -15.625,
      "loss": 25.037967681884766,
      "rewards/accuracies": 0.5,
      "rewards/chosen": -151.04458618164062,
      "rewards/margins": 0.520965576171875,
      "rewards/rejected": -151.5655517578125,
      "step": 80
    },
    {
      "epoch": 0.6177606177606177,
      "grad_norm": 845.4769875876663,
      "learning_rate": 1.993912420112756e-07,
      "logits/chosen": -0.8359375,
      "logits/rejected": -0.828125,
      "logps/chosen": -15.5,
      "logps/rejected": -15.4375,
      "loss": 24.371807098388672,
      "rewards/accuracies": 0.6000000238418579,
      "rewards/chosen": -149.18389892578125,
      "rewards/margins": 1.032012939453125,
      "rewards/rejected": -150.21591186523438,
      "step": 90
    },
    {
      "epoch": 0.6864006864006864,
      "grad_norm": 741.7784259326611,
      "learning_rate": 1.426879939713322e-07,
      "logits/chosen": -0.8125,
      "logits/rejected": -0.796875,
      "logps/chosen": -15.375,
      "logps/rejected": -15.4375,
      "loss": 24.30291748046875,
      "rewards/accuracies": 0.581250011920929,
      "rewards/chosen": -148.29949951171875,
      "rewards/margins": 1.22161865234375,
      "rewards/rejected": -149.52110290527344,
      "step": 100
    },
    {
      "epoch": 0.7550407550407551,
      "grad_norm": 891.2761904365972,
      "learning_rate": 9.212693310317479e-08,
      "logits/chosen": -0.81640625,
      "logits/rejected": -0.80859375,
      "logps/chosen": -15.4375,
      "logps/rejected": -15.4375,
      "loss": 24.666929244995117,
      "rewards/accuracies": 0.512499988079071,
      "rewards/chosen": -148.9534912109375,
      "rewards/margins": 0.953338623046875,
      "rewards/rejected": -149.90682983398438,
      "step": 110
    },
    {
      "epoch": 0.8236808236808236,
      "grad_norm": 837.0292977089434,
      "learning_rate": 5.060200865767605e-08,
      "logits/chosen": -0.77734375,
      "logits/rejected": -0.78515625,
      "logps/chosen": -15.5,
      "logps/rejected": -15.5,
      "loss": 24.336681365966797,
      "rewards/accuracies": 0.550000011920929,
      "rewards/chosen": -148.3778076171875,
      "rewards/margins": 1.705291748046875,
      "rewards/rejected": -150.08309936523438,
      "step": 120
    },
    {
      "epoch": 0.8923208923208923,
      "grad_norm": 912.1427660596082,
      "learning_rate": 2.0489971081290193e-08,
      "logits/chosen": -0.79296875,
      "logits/rejected": -0.77734375,
      "logps/chosen": -15.3125,
      "logps/rejected": -15.375,
      "loss": 24.866641998291016,
      "rewards/accuracies": 0.512499988079071,
      "rewards/chosen": -148.61953735351562,
      "rewards/margins": 0.5936431884765625,
      "rewards/rejected": -149.2131805419922,
      "step": 130
    },
    {
      "epoch": 0.960960960960961,
      "grad_norm": 1236.4699022849795,
      "learning_rate": 3.5143346177878565e-09,
      "logits/chosen": -0.77734375,
      "logits/rejected": -0.78125,
      "logps/chosen": -15.375,
      "logps/rejected": -15.375,
      "loss": 24.20675277709961,
      "rewards/accuracies": 0.5562499761581421,
      "rewards/chosen": -148.130126953125,
      "rewards/margins": 1.1678466796875,
      "rewards/rejected": -149.2979736328125,
      "step": 140
    },
    {
      "epoch": 1.0,
      "eval_logits/chosen": -0.8125,
      "eval_logits/rejected": -0.7734375,
      "eval_logps/chosen": -15.25,
      "eval_logps/rejected": -15.375,
      "eval_loss": 23.90784454345703,
      "eval_rewards/accuracies": 0.5526315569877625,
      "eval_rewards/chosen": -148.4471893310547,
      "eval_rewards/margins": 1.3118354082107544,
      "eval_rewards/rejected": -149.759033203125,
      "eval_runtime": 70.7414,
      "eval_samples_per_second": 8.538,
      "eval_steps_per_second": 1.074,
      "step": 146
    },
    {
      "epoch": 1.0,
      "step": 146,
      "total_flos": 0.0,
      "train_loss": 24.544400567877783,
      "train_runtime": 3326.8166,
      "train_samples_per_second": 5.604,
      "train_steps_per_second": 0.044
    }
  ],
  "logging_steps": 10,
  "max_steps": 146,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 1,
  "save_steps": 1000,
  "stateful_callbacks": {
    "TrainerControl": {
      "args": {
        "should_epoch_stop": false,
        "should_evaluate": false,
        "should_log": false,
        "should_save": true,
        "should_training_stop": true
      },
      "attributes": {}
    }
  },
  "total_flos": 0.0,
  "train_batch_size": 1,
  "trial_name": null,
  "trial_params": null
}