Any leads on how to fine-tune Pixtral on custom dataset for reference?

#12
by awasmohit - opened

I would like to experiment with the model after finetuning it with my custom dataset where I have noticed limited capabilities.
Apart from an online video from Trelis research on YT (he is also providing the notebook behind paywall), there is no other publically available help.

I would be grateful if anyone can share the leads under this thread.

Thanks in advance.

Unofficial Mistral Community org

Hi,

I highly recommend this repo: https://github.com/wjbmattingly/qwen2-vl-finetune-huggingface. It's very similar to fine-tune Pixtral-12B.

Let me know if you face any issues

Pixtral is th llavanext model so you can find a training script for the llava !

Hi~ Maybe this patch of code can help you fine-tune Pixtral:https://github.com/hiyouga/LLaMA-Factory/pull/5581.

Sign up or log in to comment