[Usage] How do you run the LLaVA NeXT-Vicuna 7B Baseline

[DanaOsama]

Describe the issue

Issue:
I am unable to run the LLaVA NeXT -Vicuna 7B baseline. Can I know the exact hugging face checkpoint you were using?
I am using the following: llava-hf/llava-v1.6-vicuna-7b-hf, but it is not working.
I have also followed the steps outlined by the lmms-eval repo, but to no avail. Can you please specify the exact command to use to run this baseline?

[jungle-gym-ac]

Hi, we use the official llava-1.6 checkpoint liuhaotian/llava-v1.6-vicuna-7b(huggingface link is here). You can use this model name and follow the steps outlined by the lmms-eval repo for evaluation.

By the way, compared to the original data used to train liuhaotian/llava-v1.6-vicuna-7b, the released 779K data lmms-lab/LLaVA-NeXT-Data lacks 15k user instruction data due to license issue and policy concern. To ensure a fair comparison, we train a LLaVA-NeXT model with official LLaVA-NeXT codebase on this released version of 779k data and use this model as our baseline.