trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 3.0,
  "eval_steps": 500,
  "global_step": 600,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 2.5,
      "learning_rate": 8.333333333333334e-06,
      "loss": 6.9234,
      "step": 500
    },
    {
      "epoch": 3.0,
      "step": 600,
      "total_flos": 52647771955200.0,
      "train_loss": 6.107138163248698,
      "train_runtime": 143.5946,
      "train_samples_per_second": 16.714,
      "train_steps_per_second": 4.178
    }
  ],
  "logging_steps": 500,
  "max_steps": 600,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 3,
  "save_steps": 500,
  "total_flos": 52647771955200.0,
  "train_batch_size": 4,
  "trial_name": null,
  "trial_params": null
}