More large LLM potentials needed for the community!

[dragen1860]

Dear authors:
It's very promising to witness that the stronger Mistral-7b llm models enhance the capability of video understanding. We would eager to see more potentials performed by replacing the llm with more strong models such as llama3, Yi-34b, InternLM. Specificlly, please try to evaluate some llm models such as 34b, 70b and let the community know whether it helps. Thank for such a great project.

[Andy1621]

Good idea! But current codebase is not friendly for larger LLM. We have updated VideoChat2-HD, which use large resolution for better results.