output_ids are all zeros

I would like to ask if the author has encountered a situation where output_ids are all zeros. The model weights are fine (single image inference works), and the dataset (nextqa) has been completed according to the given format.

output_ids = model.generate(
**data_dict,
do_sample=do_sample,
temperature=temperature,
max_new_tokens=256,
top_p=top_p,
top_k=top_k,
use_cache=True,
stopping_criteria=[stopping_criteria],
pad_token_id=tokenizer.eos_token_id,
)

How should I solve this issue?