trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 5.0,
  "eval_steps": 500,
  "global_step": 915,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 2.73,
      "grad_norm": NaN,
      "learning_rate": 4.9726775956284156e-05,
      "loss": 8.8735,
      "step": 500
    },
    {
      "epoch": 5.0,
      "step": 915,
      "total_flos": 9.09670937966674e+16,
      "train_loss": 4.84892471396858,
      "train_runtime": 273.84,
      "train_samples_per_second": 212.88,
      "train_steps_per_second": 3.341
    }
  ],
  "logging_steps": 500,
  "max_steps": 915,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 5,
  "save_steps": 500,
  "total_flos": 9.09670937966674e+16,
  "train_batch_size": 8,
  "trial_name": null,
  "trial_params": null
}