checkpoint-100/trainer_state.json

{
  "best_metric": 0.6628985404968262,
  "best_model_checkpoint": "./Zephyr/14-03-24-Weni-WeniGPT-2.4.1-Zephyr-7B-3-epochs-LLM_Base_2.0.3_DPO_WeniGPT DPO training-2_max_steps-267_batch_32_2024-03-14_ppid_9/checkpoint-100",
  "epoch": 1.1173184357541899,
  "eval_steps": 100,
  "global_step": 100,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 0.22,
      "grad_norm": 2.263607978820801,
      "learning_rate": 3.518518518518519e-05,
      "logits/chosen": -2.214717149734497,
      "logits/rejected": -2.2155280113220215,
      "logps/chosen": -5.710334777832031,
      "logps/rejected": -9.765484809875488,
      "loss": 0.6883,
      "rewards/accuracies": 0.04062499850988388,
      "rewards/chosen": 0.009401815943419933,
      "rewards/margins": 0.012448241002857685,
      "rewards/rejected": -0.0030464245937764645,
      "step": 20
    },
    {
      "epoch": 0.45,
      "grad_norm": 0.008143444545567036,
      "learning_rate": 4.75e-05,
      "logits/chosen": -2.2037911415100098,
      "logits/rejected": -2.2049593925476074,
      "logps/chosen": -4.532853126525879,
      "logps/rejected": -9.577239036560059,
      "loss": 0.6685,
      "rewards/accuracies": 0.05000000074505806,
      "rewards/chosen": 0.05643541365861893,
      "rewards/margins": 0.14316150546073914,
      "rewards/rejected": -0.0867261067032814,
      "step": 40
    },
    {
      "epoch": 0.67,
      "grad_norm": 0.03508400544524193,
      "learning_rate": 4.3333333333333334e-05,
      "logits/chosen": -2.1209869384765625,
      "logits/rejected": -2.1222269535064697,
      "logps/chosen": -8.109282493591309,
      "logps/rejected": -18.921493530273438,
      "loss": 0.648,
      "rewards/accuracies": 0.06875000149011612,
      "rewards/chosen": -0.08420856297016144,
      "rewards/margins": 0.552670955657959,
      "rewards/rejected": -0.636879563331604,
      "step": 60
    },
    {
      "epoch": 0.89,
      "grad_norm": 9.134880656347377e-07,
      "learning_rate": 3.9166666666666665e-05,
      "logits/chosen": -2.05387020111084,
      "logits/rejected": -2.0554463863372803,
      "logps/chosen": -10.270573616027832,
      "logps/rejected": -23.5235595703125,
      "loss": 0.6617,
      "rewards/accuracies": 0.0703125,
      "rewards/chosen": -0.29610732197761536,
      "rewards/margins": 0.937633216381073,
      "rewards/rejected": -1.2337405681610107,
      "step": 80
    },
    {
      "epoch": 1.12,
      "grad_norm": 0.011362694203853607,
      "learning_rate": 3.5e-05,
      "logits/chosen": -2.0597641468048096,
      "logits/rejected": -2.061570167541504,
      "logps/chosen": -8.65438175201416,
      "logps/rejected": -23.752864837646484,
      "loss": 0.645,
      "rewards/accuracies": 0.0703125,
      "rewards/chosen": -0.09740939736366272,
      "rewards/margins": 1.0688247680664062,
      "rewards/rejected": -1.1662342548370361,
      "step": 100
    },
    {
      "epoch": 1.12,
      "eval_logits/chosen": -2.0584123134613037,
      "eval_logits/rejected": -2.058645725250244,
      "eval_logps/chosen": -5.4038591384887695,
      "eval_logps/rejected": -15.805567741394043,
      "eval_loss": 0.6628985404968262,
      "eval_rewards/accuracies": 0.046875,
      "eval_rewards/chosen": -0.038359977304935455,
      "eval_rewards/margins": 0.7473276853561401,
      "eval_rewards/rejected": -0.7856876254081726,
      "eval_runtime": 184.5849,
      "eval_samples_per_second": 1.723,
      "eval_steps_per_second": 0.217,
      "step": 100
    }
  ],
  "logging_steps": 20,
  "max_steps": 267,
  "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
}