Can we train a VLM to ๐ฉ๐ซ๐๐๐๐ซ?
This is now possible, thanks to the new TRL/DPO support for VLMs! ๐
As an example, we've trained a model to reduce hallucinations.
Check out:
๐ฐ Blog post: https://huggingface.co/blog/dpo_vlm
๐ TRL: https://github.com/huggingface/trl
Thanks to
@mervenoyann
,
@vwxyzjn
and
@krasul
who helped me with this work!
This is now possible, thanks to the new TRL/DPO support for VLMs! ๐
As an example, we've trained a model to reduce hallucinations.
Check out:
๐ฐ Blog post: https://huggingface.co/blog/dpo_vlm
๐ TRL: https://github.com/huggingface/trl
Thanks to
@mervenoyann
,
@vwxyzjn
and
@krasul
who helped me with this work!