finetune.py

model_args = {
    "reprocess_input_data": True,
    "overwrite_output_dir": True,
    "max_seq_length": 128,
    "train_batch_size": 16,
    "num_train_epochs": 25,
    "num_beams": None,
    "do_sample": True,
    "max_length": 20,
    "top_k": 50,
    "top_p": 0.95,
    "use_multiprocessing": False,
    "save_steps": -1,
    "save_eval_checkpoints": True,
    "evaluate_during_training": True,
    "evaluate_during_training_verbose": True,
    "num_return_sequences": 5
}