{
  "best_global_step": 146,
  "best_metric": 0.09535243362188339,
  "best_model_checkpoint": "/NHNHOME/WORKSPACE/26msit001_A/mnpo/revision_qwen3_8b/full_iter1/train/htmnpo_skywork_s42_nhn-b200-q3-seq-07101430-htmnpo/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": 0.06866952789699571,
      "grad_norm": 658.2609913586693,
      "ht_mnpo/logit": -0.07977066189050674,
      "ht_mnpo/residual": -0.08727066218852997,
      "ht_mnpo/residual_abs": 0.5136820673942566,
      "ht_mnpo/target": 0.007499999366700649,
      "ht_mnpo/target_abs": 0.007499999366700649,
      "learning_rate": 3e-07,
      "logits/chosen": -0.81640625,
      "logits/rejected": -0.81640625,
      "logps/chosen": -15.9375,
      "logps/rejected": -15.875,
      "loss": 1.032188057899475,
      "rewards/accuracies": 0.4749999940395355,
      "rewards/chosen": -154.9237518310547,
      "rewards/margins": -0.7977066040039062,
      "rewards/rejected": -154.1260528564453,
      "step": 10
    },
    {
      "epoch": 0.13733905579399142,
      "grad_norm": 486.85604769071176,
      "ht_mnpo/logit": 0.07697143405675888,
      "ht_mnpo/residual": 0.06947144120931625,
      "ht_mnpo/residual_abs": 0.46500977873802185,
      "ht_mnpo/target": 0.007499999366700649,
      "ht_mnpo/target_abs": 0.007499999366700649,
      "learning_rate": 4.988506452457066e-07,
      "logits/chosen": -0.80859375,
      "logits/rejected": -0.83984375,
      "logps/chosen": -15.875,
      "logps/rejected": -15.875,
      "loss": 0.5586695075035095,
      "rewards/accuracies": 0.543749988079071,
      "rewards/chosen": -153.60064697265625,
      "rewards/margins": 0.76971435546875,
      "rewards/rejected": -154.370361328125,
      "step": 20
    },
    {
      "epoch": 0.20600858369098712,
      "grad_norm": 279.6489736984974,
      "ht_mnpo/logit": 0.08701781928539276,
      "ht_mnpo/residual": 0.07951782643795013,
      "ht_mnpo/residual_abs": 0.48369044065475464,
      "ht_mnpo/target": 0.007499999366700649,
      "ht_mnpo/target_abs": 0.007499999366700649,
      "learning_rate": 4.860414659112948e-07,
      "logits/chosen": -0.8359375,
      "logits/rejected": -0.8359375,
      "logps/chosen": -15.75,
      "logps/rejected": -15.75,
      "loss": 0.6932045221328735,
      "rewards/accuracies": 0.543749988079071,
      "rewards/chosen": -152.07644653320312,
      "rewards/margins": 0.87017822265625,
      "rewards/rejected": -152.94662475585938,
      "step": 30
    },
    {
      "epoch": 0.27467811158798283,
      "grad_norm": 405.5407274927664,
      "ht_mnpo/logit": 0.11707458645105362,
      "ht_mnpo/residual": 0.1095745787024498,
      "ht_mnpo/residual_abs": 0.40353816747665405,
      "ht_mnpo/target": 0.007499999366700649,
      "ht_mnpo/target_abs": 0.007499999366700649,
      "learning_rate": 4.597220474379125e-07,
      "logits/chosen": -0.82421875,
      "logits/rejected": -0.82421875,
      "logps/chosen": -15.5,
      "logps/rejected": -15.5,
      "loss": 0.6164440512657166,
      "rewards/accuracies": 0.5375000238418579,
      "rewards/chosen": -149.45849609375,
      "rewards/margins": 1.170745849609375,
      "rewards/rejected": -150.62924194335938,
      "step": 40
    },
    {
      "epoch": 0.34334763948497854,
      "grad_norm": 173.47778729238564,
      "ht_mnpo/logit": 0.09282531589269638,
      "ht_mnpo/residual": 0.08532532304525375,
      "ht_mnpo/residual_abs": 0.3633372485637665,
      "ht_mnpo/target": 0.007499999366700649,
      "ht_mnpo/target_abs": 0.007499999366700649,
      "learning_rate": 4.2139882699413613e-07,
      "logits/chosen": -0.89453125,
      "logits/rejected": -0.921875,
      "logps/chosen": -14.9375,
      "logps/rejected": -14.9375,
      "loss": 0.38052481412887573,
      "rewards/accuracies": 0.6000000238418579,
      "rewards/chosen": -144.1920166015625,
      "rewards/margins": 0.928253173828125,
      "rewards/rejected": -145.12026977539062,
      "step": 50
    },
    {
      "epoch": 0.41201716738197425,
      "grad_norm": 220.4191286143336,
      "ht_mnpo/logit": 0.15768662095069885,
      "ht_mnpo/residual": 0.15018662810325623,
      "ht_mnpo/residual_abs": 0.4016141891479492,
      "ht_mnpo/target": 0.007499999366700649,
      "ht_mnpo/target_abs": 0.007499999366700649,
      "learning_rate": 3.7326529999303633e-07,
      "logits/chosen": -0.89453125,
      "logits/rejected": -0.91796875,
      "logps/chosen": -14.5625,
      "logps/rejected": -14.625,
      "loss": 0.8031173944473267,
      "rewards/accuracies": 0.5687500238418579,
      "rewards/chosen": -140.22152709960938,
      "rewards/margins": 1.5768661499023438,
      "rewards/rejected": -141.7983856201172,
      "step": 60
    },
    {
      "epoch": 0.48068669527896996,
      "grad_norm": 226.8002386346793,
      "ht_mnpo/logit": 0.15667113661766052,
      "ht_mnpo/residual": 0.1491711437702179,
      "ht_mnpo/residual_abs": 0.4558327794075012,
      "ht_mnpo/target": 0.007499999366700649,
      "ht_mnpo/target_abs": 0.007499999366700649,
      "learning_rate": 3.1807647161082797e-07,
      "logits/chosen": -0.890625,
      "logits/rejected": -0.8828125,
      "logps/chosen": -14.3125,
      "logps/rejected": -14.3125,
      "loss": 1.1153709888458252,
      "rewards/accuracies": 0.512499988079071,
      "rewards/chosen": -136.7957305908203,
      "rewards/margins": 1.56671142578125,
      "rewards/rejected": -138.36244201660156,
      "step": 70
    },
    {
      "epoch": 0.5493562231759657,
      "grad_norm": 188.6296821878058,
      "ht_mnpo/logit": 0.040885161608457565,
      "ht_mnpo/residual": 0.03338516503572464,
      "ht_mnpo/residual_abs": 0.35882776975631714,
      "ht_mnpo/target": 0.007499999366700649,
      "ht_mnpo/target_abs": 0.007499999366700649,
      "learning_rate": 2.5899116932879534e-07,
      "logits/chosen": -0.83203125,
      "logits/rejected": -0.87109375,
      "logps/chosen": -14.1875,
      "logps/rejected": -14.1875,
      "loss": 0.46190714836120605,
      "rewards/accuracies": 0.4937500059604645,
      "rewards/chosen": -136.6307373046875,
      "rewards/margins": 0.40885162353515625,
      "rewards/rejected": -137.03958129882812,
      "step": 80
    },
    {
      "epoch": 0.6180257510729614,
      "grad_norm": 159.63381058216007,
      "ht_mnpo/logit": 0.2624649107456207,
      "ht_mnpo/residual": 0.2549648880958557,
      "ht_mnpo/residual_abs": 0.4282897412776947,
      "ht_mnpo/target": 0.007499999366700649,
      "ht_mnpo/target_abs": 0.007499999366700649,
      "learning_rate": 1.993912420112756e-07,
      "logits/chosen": -0.83984375,
      "logits/rejected": -0.84375,
      "logps/chosen": -14.0625,
      "logps/rejected": -14.125,
      "loss": 0.8312055468559265,
      "rewards/accuracies": 0.6000000238418579,
      "rewards/chosen": -134.35748291015625,
      "rewards/margins": 2.6246490478515625,
      "rewards/rejected": -136.9821319580078,
      "step": 90
    },
    {
      "epoch": 0.6866952789699571,
      "grad_norm": 217.2628805931693,
      "ht_mnpo/logit": 0.059761811047792435,
      "ht_mnpo/residual": 0.05226181820034981,
      "ht_mnpo/residual_abs": 0.303397536277771,
      "ht_mnpo/target": 0.007499999366700649,
      "ht_mnpo/target_abs": 0.007499999366700649,
      "learning_rate": 1.426879939713322e-07,
      "logits/chosen": -0.86328125,
      "logits/rejected": -0.875,
      "logps/chosen": -13.875,
      "logps/rejected": -13.875,
      "loss": 0.22050431370735168,
      "rewards/accuracies": 0.518750011920929,
      "rewards/chosen": -134.1347198486328,
      "rewards/margins": 0.5976181030273438,
      "rewards/rejected": -134.73233032226562,
      "step": 100
    },
    {
      "epoch": 0.7553648068669528,
      "grad_norm": 307.6598272099915,
      "ht_mnpo/logit": 0.08397064357995987,
      "ht_mnpo/residual": 0.07647065818309784,
      "ht_mnpo/residual_abs": 0.3651581406593323,
      "ht_mnpo/target": 0.007499999366700649,
      "ht_mnpo/target_abs": 0.007499999366700649,
      "learning_rate": 9.212693310317479e-08,
      "logits/chosen": -0.83203125,
      "logits/rejected": -0.84765625,
      "logps/chosen": -13.9375,
      "logps/rejected": -13.9375,
      "loss": 0.46880072355270386,
      "rewards/accuracies": 0.5562499761581421,
      "rewards/chosen": -133.74819946289062,
      "rewards/margins": 0.8397064208984375,
      "rewards/rejected": -134.58790588378906,
      "step": 110
    },
    {
      "epoch": 0.8240343347639485,
      "grad_norm": 101.02657333091942,
      "ht_mnpo/logit": 0.02447204664349556,
      "ht_mnpo/residual": 0.016972053796052933,
      "ht_mnpo/residual_abs": 0.24492737650871277,
      "ht_mnpo/target": 0.007499999366700649,
      "ht_mnpo/target_abs": 0.007499999366700649,
      "learning_rate": 5.060200865767605e-08,
      "logits/chosen": -0.80859375,
      "logits/rejected": -0.8359375,
      "logps/chosen": -13.8125,
      "logps/rejected": -13.875,
      "loss": 0.1213923841714859,
      "rewards/accuracies": 0.5625,
      "rewards/chosen": -133.80438232421875,
      "rewards/margins": 0.244720458984375,
      "rewards/rejected": -134.04910278320312,
      "step": 120
    },
    {
      "epoch": 0.8927038626609443,
      "grad_norm": 62.457306149658216,
      "ht_mnpo/logit": 0.18063659965991974,
      "ht_mnpo/residual": 0.1731366068124771,
      "ht_mnpo/residual_abs": 0.3555470108985901,
      "ht_mnpo/target": 0.007499999366700649,
      "ht_mnpo/target_abs": 0.007499999366700649,
      "learning_rate": 2.0489971081290193e-08,
      "logits/chosen": -0.78125,
      "logits/rejected": -0.80859375,
      "logps/chosen": -13.875,
      "logps/rejected": -13.9375,
      "loss": 0.5861908197402954,
      "rewards/accuracies": 0.5874999761581421,
      "rewards/chosen": -133.495361328125,
      "rewards/margins": 1.806365966796875,
      "rewards/rejected": -135.30172729492188,
      "step": 130
    },
    {
      "epoch": 0.9613733905579399,
      "grad_norm": 59.652930086354274,
      "ht_mnpo/logit": 0.009463501162827015,
      "ht_mnpo/residual": 0.001963504124432802,
      "ht_mnpo/residual_abs": 0.3243860602378845,
      "ht_mnpo/target": 0.007499999366700649,
      "ht_mnpo/target_abs": 0.007499999366700649,
      "learning_rate": 3.5143346177878565e-09,
      "logits/chosen": -0.79296875,
      "logits/rejected": -0.83203125,
      "logps/chosen": -13.875,
      "logps/rejected": -13.9375,
      "loss": 0.5267720818519592,
      "rewards/accuracies": 0.518750011920929,
      "rewards/chosen": -134.2991943359375,
      "rewards/margins": 0.094635009765625,
      "rewards/rejected": -134.39382934570312,
      "step": 140
    },
    {
      "epoch": 1.0,
      "eval_ht_mnpo/logit": 0.0946219190955162,
      "eval_ht_mnpo/residual": 0.08712193369865417,
      "eval_ht_mnpo/residual_abs": 0.23038248717784882,
      "eval_ht_mnpo/target": 0.007499998901039362,
      "eval_ht_mnpo/target_abs": 0.007499998901039362,
      "eval_logits/chosen": -0.78515625,
      "eval_logits/rejected": -0.76953125,
      "eval_logps/chosen": -13.8125,
      "eval_logps/rejected": -13.875,
      "eval_loss": 0.09535243362188339,
      "eval_rewards/accuracies": 0.6578947305679321,
      "eval_rewards/chosen": -134.2300262451172,
      "eval_rewards/margins": 0.9462206959724426,
      "eval_rewards/rejected": -135.17623901367188,
      "eval_runtime": 70.1838,
      "eval_samples_per_second": 8.606,
      "eval_steps_per_second": 1.083,
      "step": 146
    },
    {
      "epoch": 1.0,
      "step": 146,
      "total_flos": 0.0,
      "train_loss": 0.5071899907229698,
      "train_runtime": 3280.3544,
      "train_samples_per_second": 5.682,
      "train_steps_per_second": 0.045
    }
  ],
  "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
}