training_state.json

{
  "iter_num": 1,
  "local_iter_num": 1,
  "best_val_loss": 8.323423385620117,
  "running_mfu": -1.0,
  "lr": 5e-07,
  "epoch": 0,
  "step": 1
}