trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 0.999973032011003,
  "eval_steps": 500,
  "global_step": 18540,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 0.5393597799412098,
      "grad_norm": 0.5843396186828613,
      "learning_rate": 2.192383594655041e-05,
      "loss": 0.4481,
      "step": 10000
    },
    {
      "epoch": 0.999973032011003,
      "step": 18540,
      "total_flos": 1.291616631271981e+18,
      "train_loss": 0.4373928493923611,
      "train_runtime": 30548.4489,
      "train_samples_per_second": 2.428,
      "train_steps_per_second": 0.607
    }
  ],
  "logging_steps": 10000,
  "max_steps": 18540,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 1,
  "save_steps": 500,
  "total_flos": 1.291616631271981e+18,
  "train_batch_size": 2,
  "trial_name": null,
  "trial_params": null
}