trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 12.966233766233767,
  "eval_steps": 500.0,
  "global_step": 624,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 10.38961038961039,
      "grad_norm": 1.4692922830581665,
      "learning_rate": 9.935897435897435e-06,
      "loss": 2.444,
      "step": 500
    },
    {
      "epoch": 12.966233766233767,
      "step": 624,
      "total_flos": 2608739647488000.0,
      "train_loss": 2.33007384569217,
      "train_runtime": 838.3975,
      "train_samples_per_second": 11.939,
      "train_steps_per_second": 0.744
    }
  ],
  "logging_steps": 500,
  "max_steps": 624,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 13,
  "save_steps": 500,
  "total_flos": 2608739647488000.0,
  "train_batch_size": 2,
  "trial_name": null,
  "trial_params": null
}