trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 3.0,
  "eval_steps": 500,
  "global_step": 3,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 1.0,
      "grad_norm": 17.48065334361642,
      "learning_rate": 0.0,
      "loss": 0.7844,
      "step": 1
    },
    {
      "epoch": 2.0,
      "grad_norm": 8.642240962023566,
      "learning_rate": 1e-05,
      "loss": 0.238,
      "step": 2
    },
    {
      "epoch": 3.0,
      "grad_norm": 8.688956609398945,
      "learning_rate": 1e-05,
      "loss": 0.2293,
      "step": 3
    },
    {
      "epoch": 3.0,
      "step": 3,
      "total_flos": 256605511680.0,
      "train_loss": 0.4172290066878001,
      "train_runtime": 67.7434,
      "train_samples_per_second": 0.443,
      "train_steps_per_second": 0.044
    }
  ],
  "logging_steps": 1.0,
  "max_steps": 3,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 3,
  "save_steps": 700,
  "total_flos": 256605511680.0,
  "train_batch_size": 4,
  "trial_name": null,
  "trial_params": null
}