on
    >>> with torch.no_grad():
    ...     generated_ids = model.generate(**inputs)

    >>> generated_text = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
    ```
a_