Support of evaluation & training of LLaVa-like models

#3
by Haon-Chen - opened

Could you add codes that support the training and evaluation of LLaVa-like models? For example, nvidia/MM-Embed.

TIGER-Lab org

@Haon-Chen
Thanks for your interest! Yes, we are working on it and we will release some model based on LLaVA-NeXT soon.

TIGER-Lab org

@Haon-Chen
A quick update here, we have released a Llava-Next based VLM2Vec here: https://huggingface.co/TIGER-Lab/VLM2Vec-LLaVa-Next
This is our current best version of VLM2Vec when evaluating on MMEB-eval and zero-shot Flickr30K.
We will continue working on supporting additional VLM backbones.

ziyjiang changed discussion status to closed

Sign up or log in to comment