LLaMA-VID icon indicating copy to clipboard operation
LLaMA-VID copied to clipboard

Published 20 hours ago • verified publisher icon dvlab-research

Cannot reproduce VideoChatGPT generative performance results

Open xsgldhy opened this issue 6 months ago • 2 comments

Thank you for your contribution!

Hello, I'm trying to reproduce the evaluation scores for generative performance in the VideoChatGPT evaluation of model EVA-G & LLaVA1.5-VideoChatGPT-Instruct 7B. I have downloaded your codebase; aside from adjusting the video path, I have also changed the name of your LlavaConfig from "llava" to "llama_vid," because the former causes a conflict with my transformers package (version 4.41.2). All the other parts remain the same. image image image

Reproduced result is shown below: image

xsgldhy avatar Jul 30 '24 02:07 xsgldhy