Could you provide evaluation code?

[JerryPW]

Could you please provide a complete evaluation script to explain how the model is being evaluated? The inference_chat.py you provided seems to only support inference for a single sample and doesn't handle large-scale batch evaluation. Also, I'm not sure whether it's an issue with my environment or your code, but after downloading the latest model from Hugging Face, I noticed that only the first inference produces a valid result, while all subsequent results are just "!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!", which is quite strange. I would appreciate it if you could thoroughly check the code or let me know what might be causing this issue.

[jkhu29]

Or, you can try another repo Omni-VIew, which supports unified understanding and generation on multi-views. It works well on the VSI-Bench.