trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 10.0,
  "eval_steps": 10,
  "global_step": 10,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 10.0,
      "learning_rate": 4e-05,
      "loss": 0.3826,
      "step": 10
    },
    {
      "epoch": 10.0,
      "eval_loss": 4.4464287757873535,
      "eval_runtime": 0.0176,
      "eval_samples_per_second": 56.724,
      "eval_steps_per_second": 56.724,
      "step": 10
    }
  ],
  "logging_steps": 10,
  "max_steps": 50,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 50,
  "save_steps": 10,
  "total_flos": 6571230957720.0,
  "train_batch_size": 16,
  "trial_name": null,
  "trial_params": null
}