trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 4.0,
  "eval_steps": 40,
  "global_step": 20,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 2.0,
      "grad_norm": 0.5412699580192566,
      "learning_rate": 0.0001,
      "loss": 0.344,
      "step": 10
    },
    {
      "epoch": 4.0,
      "grad_norm": 0.285557359457016,
      "learning_rate": 0.0,
      "loss": 0.1246,
      "step": 20
    },
    {
      "epoch": 4.0,
      "step": 20,
      "total_flos": 2.3867532801294336e+16,
      "train_loss": 0.23429088592529296,
      "train_runtime": 606.0988,
      "train_samples_per_second": 0.508,
      "train_steps_per_second": 0.033
    }
  ],
  "logging_steps": 10,
  "max_steps": 20,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 4,
  "save_steps": 40,
  "total_flos": 2.3867532801294336e+16,
  "train_batch_size": 16,
  "trial_name": null,
  "trial_params": null
}