Using the fine-tuned model for video inference, the output is null

[aFishCalledSakana]

I used a fine-tuned model for video inference, but the output was null. However, there was no problem with text inference. Why is this happening? How can I solve this problem?

[aFishCalledSakana]

以上第一张图片是官方的模型输出的图片编码结果，第二张是自己微调的模型图片编码的结果，输出为nan，导致output_id=0