trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 4.0,
  "eval_steps": 100,
  "global_step": 12,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 3.33,
      "grad_norm": 0.951649010181427,
      "learning_rate": 3.3333333333333335e-05,
      "loss": 0.5381,
      "step": 10
    },
    {
      "epoch": 4.0,
      "step": 12,
      "total_flos": 1.1468814462959616e+16,
      "train_loss": 0.49794552226861316,
      "train_runtime": 44.8816,
      "train_samples_per_second": 3.298,
      "train_steps_per_second": 0.267
    }
  ],
  "logging_steps": 10,
  "max_steps": 12,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 4,
  "save_steps": 100,
  "total_flos": 1.1468814462959616e+16,
  "train_batch_size": 16,
  "trial_name": null,
  "trial_params": null
}