checkpoint-100/trainer_state.json

{
  "best_metric": 0.12640328705310822,
  "best_model_checkpoint": "./Mistral/16-03-24-Weni-ZeroShot-3.4.16-Mistral-7b-DPO-1.0.0_ZeroShot DPO Training a improved dataset and with learning rate 40x smaller than SFT-2_max_steps-144_batch_32_2024-03-16_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": 26.23642921447754,
      "learning_rate": 2e-05,
      "logits/chosen": -1.1930553913116455,
      "logits/rejected": -1.1628875732421875,
      "logps/chosen": -21.34621810913086,
      "logps/rejected": -19.047561645507812,
      "loss": 0.5766,
      "rewards/accuracies": 0.520312488079071,
      "rewards/chosen": 0.26292288303375244,
      "rewards/margins": 0.6182515025138855,
      "rewards/rejected": -0.35532867908477783,
      "step": 20
    },
    {
      "epoch": 0.82,
      "grad_norm": 17.913877487182617,
      "learning_rate": 1.689922480620155e-05,
      "logits/chosen": -1.13822603225708,
      "logits/rejected": -1.0975478887557983,
      "logps/chosen": -19.194286346435547,
      "logps/rejected": -20.591718673706055,
      "loss": 0.3335,
      "rewards/accuracies": 0.8687499761581421,
      "rewards/chosen": 1.3181432485580444,
      "rewards/margins": 3.1427412033081055,
      "rewards/rejected": -1.824597716331482,
      "step": 40
    },
    {
      "epoch": 1.02,
      "eval_logits/chosen": -1.100258231163025,
      "eval_logits/rejected": -1.0552455186843872,
      "eval_logps/chosen": -20.249515533447266,
      "eval_logps/rejected": -23.880615234375,
      "eval_loss": 0.1760517805814743,
      "eval_rewards/accuracies": 0.9147727489471436,
      "eval_rewards/chosen": 0.4012405574321747,
      "eval_rewards/margins": 4.824998378753662,
      "eval_rewards/rejected": -4.423757553100586,
      "eval_runtime": 82.5821,
      "eval_samples_per_second": 2.107,
      "eval_steps_per_second": 0.266,
      "step": 50
    },
    {
      "epoch": 1.22,
      "grad_norm": 13.158234596252441,
      "learning_rate": 1.3798449612403102e-05,
      "logits/chosen": -1.141601324081421,
      "logits/rejected": -1.1046007871627808,
      "logps/chosen": -19.948759078979492,
      "logps/rejected": -22.05379867553711,
      "loss": 0.2502,
      "rewards/accuracies": 0.8687499761581421,
      "rewards/chosen": 0.9041095972061157,
      "rewards/margins": 3.8345859050750732,
      "rewards/rejected": -2.930476665496826,
      "step": 60
    },
    {
      "epoch": 1.63,
      "grad_norm": 4.365915775299072,
      "learning_rate": 1.0697674418604651e-05,
      "logits/chosen": -1.1466914415359497,
      "logits/rejected": -1.1094379425048828,
      "logps/chosen": -19.571247100830078,
      "logps/rejected": -21.93134880065918,
      "loss": 0.1835,
      "rewards/accuracies": 0.926562488079071,
      "rewards/chosen": 1.109068751335144,
      "rewards/margins": 3.9569854736328125,
      "rewards/rejected": -2.847917079925537,
      "step": 80
    },
    {
      "epoch": 2.04,
      "grad_norm": 7.621004104614258,
      "learning_rate": 7.596899224806202e-06,
      "logits/chosen": -1.1677355766296387,
      "logits/rejected": -1.133095622062683,
      "logps/chosen": -20.18477439880371,
      "logps/rejected": -22.280181884765625,
      "loss": 0.1615,
      "rewards/accuracies": 0.9156249761581421,
      "rewards/chosen": 1.1184756755828857,
      "rewards/margins": 4.2123236656188965,
      "rewards/rejected": -3.0938479900360107,
      "step": 100
    },
    {
      "epoch": 2.04,
      "eval_logits/chosen": -1.1095269918441772,
      "eval_logits/rejected": -1.0665117502212524,
      "eval_logps/chosen": -19.742412567138672,
      "eval_logps/rejected": -23.139616012573242,
      "eval_loss": 0.12640328705310822,
      "eval_rewards/accuracies": 0.9318181872367859,
      "eval_rewards/chosen": 0.8069248199462891,
      "eval_rewards/margins": 4.637885570526123,
      "eval_rewards/rejected": -3.830960750579834,
      "eval_runtime": 82.6113,
      "eval_samples_per_second": 2.106,
      "eval_steps_per_second": 0.266,
      "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
}