best_trial.json

{
  "learning_rate": 3e-05,
  "train_batch_size": 8,
  "num_train_epochs": 7
}