Feature request

Is it possible to have GRPO for VLM models soon?
Or does anyone know any alternative library for that?
Thanks!

Motivation

It will help to create multimodal R1 using the same GRPO trainer.

Your contribution

N/A