trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 1.0,
  "eval_steps": 100,
  "global_step": 2,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 0.5,
      "grad_norm": 0.58984375,
      "learning_rate": 5e-06,
      "logits/chosen": -3.1927170753479004,
      "logits/rejected": -3.1942551136016846,
      "logps/chosen": -293.9735107421875,
      "logps/rejected": -260.54400634765625,
      "loss": 0.5,
      "rewards/accuracies": 0.34375,
      "rewards/chosen": -0.0009545064531266689,
      "rewards/margins": -0.0002557558473199606,
      "rewards/rejected": -0.0006987505475990474,
      "step": 1
    },
    {
      "epoch": 1.0,
      "step": 2,
      "total_flos": 0.0,
      "train_loss": 0.5000961124897003,
      "train_runtime": 87.0205,
      "train_samples_per_second": 0.701,
      "train_steps_per_second": 0.023
    }
  ],
  "logging_steps": 10,
  "max_steps": 2,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 1,
  "save_steps": 100,
  "total_flos": 0.0,
  "train_batch_size": 4,
  "trial_name": null,
  "trial_params": null
}