trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 1.92,
  "eval_steps": 500,
  "global_step": 36,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 1.3333333333333333,
      "grad_norm": 0.9292863011360168,
      "learning_rate": 4.230769230769231e-05,
      "loss": 0.4778,
      "step": 25
    },
    {
      "epoch": 1.92,
      "step": 36,
      "total_flos": 3.444055882369344e+16,
      "train_loss": 0.35278993679417503,
      "train_runtime": 5731.2572,
      "train_samples_per_second": 0.105,
      "train_steps_per_second": 0.006
    }
  ],
  "logging_steps": 25,
  "max_steps": 36,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 2,
  "save_steps": 250,
  "total_flos": 3.444055882369344e+16,
  "train_batch_size": 2,
  "trial_name": null,
  "trial_params": null
}