checkpoint-100/trainer_state.json

{
  "best_metric": 0.2397383749485016,
  "best_model_checkpoint": "./Mistral/15-03-24-Weni-ZeroShot-3.4.12-Mistral-7b-DPO-1.0.0_ZeroShot DPO Training a improved dataset and best hyperparameters found so far-2_max_steps-144_batch_32_2024-03-15_ppid_9/checkpoint-100",
  "epoch": 2.0408163265306123,
  "eval_steps": 50,
  "global_step": 100,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 0.41,
      "grad_norm": 5.192111492156982,
      "learning_rate": 0.0001937984496124031,
      "logits/chosen": -0.9390562176704407,
      "logits/rejected": -0.9053889513015747,
      "logps/chosen": -21.715065002441406,
      "logps/rejected": -29.212356567382812,
      "loss": 0.4668,
      "rewards/accuracies": 0.659375011920929,
      "rewards/chosen": -0.06483893096446991,
      "rewards/margins": 1.010897159576416,
      "rewards/rejected": -1.075736165046692,
      "step": 20
    },
    {
      "epoch": 0.82,
      "grad_norm": 1.8516901731491089,
      "learning_rate": 0.00016279069767441862,
      "logits/chosen": -0.5763565301895142,
      "logits/rejected": -0.4915798306465149,
      "logps/chosen": -21.638835906982422,
      "logps/rejected": -44.50564956665039,
      "loss": 0.2617,
      "rewards/accuracies": 0.8968750238418579,
      "rewards/chosen": -0.01356862299144268,
      "rewards/margins": 2.59334135055542,
      "rewards/rejected": -2.606910228729248,
      "step": 40
    },
    {
      "epoch": 1.02,
      "eval_logits/chosen": -0.6646511554718018,
      "eval_logits/rejected": -0.5983995795249939,
      "eval_logps/chosen": -21.368825912475586,
      "eval_logps/rejected": -46.24407196044922,
      "eval_loss": 0.2652554214000702,
      "eval_rewards/accuracies": 0.8522727489471436,
      "eval_rewards/chosen": -0.04463845491409302,
      "eval_rewards/margins": 2.807713747024536,
      "eval_rewards/rejected": -2.8523521423339844,
      "eval_runtime": 81.9688,
      "eval_samples_per_second": 2.123,
      "eval_steps_per_second": 0.268,
      "step": 50
    },
    {
      "epoch": 1.22,
      "grad_norm": 3.232189178466797,
      "learning_rate": 0.0001317829457364341,
      "logits/chosen": -0.6450687050819397,
      "logits/rejected": -0.5619121193885803,
      "logps/chosen": -20.606281280517578,
      "logps/rejected": -48.98772430419922,
      "loss": 0.2055,
      "rewards/accuracies": 0.909375011920929,
      "rewards/chosen": 0.04113447293639183,
      "rewards/margins": 3.0845394134521484,
      "rewards/rejected": -3.0434048175811768,
      "step": 60
    },
    {
      "epoch": 1.63,
      "grad_norm": 2.522237777709961,
      "learning_rate": 0.00010077519379844962,
      "logits/chosen": -0.4938630163669586,
      "logits/rejected": -0.43368759751319885,
      "logps/chosen": -17.037982940673828,
      "logps/rejected": -50.90436553955078,
      "loss": 0.1921,
      "rewards/accuracies": 0.917187511920929,
      "rewards/chosen": 0.40310320258140564,
      "rewards/margins": 3.6394691467285156,
      "rewards/rejected": -3.236365795135498,
      "step": 80
    },
    {
      "epoch": 2.04,
      "grad_norm": 2.4511005878448486,
      "learning_rate": 7.131782945736435e-05,
      "logits/chosen": -0.4601399898529053,
      "logits/rejected": -0.41903334856033325,
      "logps/chosen": -22.506694793701172,
      "logps/rejected": -49.0537223815918,
      "loss": 0.2375,
      "rewards/accuracies": 0.893750011920929,
      "rewards/chosen": -0.1375463902950287,
      "rewards/margins": 2.9313812255859375,
      "rewards/rejected": -3.068927764892578,
      "step": 100
    },
    {
      "epoch": 2.04,
      "eval_logits/chosen": -0.4721185564994812,
      "eval_logits/rejected": -0.438131719827652,
      "eval_logps/chosen": -19.791969299316406,
      "eval_logps/rejected": -41.700191497802734,
      "eval_loss": 0.2397383749485016,
      "eval_rewards/accuracies": 0.8636363744735718,
      "eval_rewards/chosen": 0.11304693669080734,
      "eval_rewards/margins": 2.5110111236572266,
      "eval_rewards/rejected": -2.3979642391204834,
      "eval_runtime": 81.8963,
      "eval_samples_per_second": 2.125,
      "eval_steps_per_second": 0.269,
      "step": 100
    }
  ],
  "logging_steps": 20,
  "max_steps": 144,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 3,
  "save_steps": 100,
  "total_flos": 0.0,
  "train_batch_size": 8,
  "trial_name": null,
  "trial_params": null
}