trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 20.0,
  "eval_steps": 500,
  "global_step": 680,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 14.71,
      "learning_rate": 1.323529411764706e-05,
      "loss": 2.8344,
      "step": 500
    },
    {
      "epoch": 20.0,
      "step": 680,
      "total_flos": 1.9265685892890624e+17,
      "train_loss": 2.621326805563534,
      "train_runtime": 1247.0654,
      "train_samples_per_second": 8.724,
      "train_steps_per_second": 0.545
    }
  ],
  "logging_steps": 500,
  "max_steps": 680,
  "num_train_epochs": 20,
  "save_steps": 200,
  "total_flos": 1.9265685892890624e+17,
  "trial_name": null,
  "trial_params": null
}