trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 3.0,
  "eval_steps": 500,
  "global_step": 72,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 3.0,
      "step": 72,
      "total_flos": 259657478920224.0,
      "train_loss": 4.278300391303168,
      "train_runtime": 734.7358,
      "train_samples_per_second": 6.076,
      "train_steps_per_second": 0.098
    }
  ],
  "logging_steps": 500,
  "max_steps": 72,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 3,
  "save_steps": 500,
  "total_flos": 259657478920224.0,
  "train_batch_size": 64,
  "trial_name": null,
  "trial_params": null
}