Progress in DPO traininer for LLaVA Next and LLaVA OV

[chenweize1998]

Feature request

It appears that the current DPO trainer doesn’t yet support LLaVA Next and LLaVA OV. According to #1784 , LLaVA Next is planned to be supported, but I’m curious about the current status or any general guidelines for adding support for these two models. Perhaps I can assist with this.

Motivation

LLaVA Next and LLaVA OV are widely adopted VLMs, so I expect to see support for the DPO trainer on them.

Your contribution

I might be able to help with integrating support for these models. I guess it's the anyres part that makes the integration a little bit tricky?