Sinoué GAD
Sinoué GAD
> Hi @GAD-cell by the way just came across your amazing repo after you posted it on Reddit. Would you be interested in integrating your VLM GRPO implementation with Unsloth?...
> > > Hi [@GAD-cell](https://github.com/GAD-cell) by the way just came across your amazing repo after you posted it on Reddit. Would you be interested in integrating your VLM GRPO implementation...
Hey @nph4rd , very cool implementation, thank you for sharing !
Hey, I'm also really interested in an implementation of GRPO for VLMs. TRL doesn't support multimodal for GRPO yet, so by extension, unsloth doesn't either. Since I really needed it,...
> @GAD-cell @danielhanchen I just tested this notebook on colab using both a float16 only GPU ( a T4) and a bfloat16 capable GPU (A100). The notebook failed on both...
Hey @rolandtannous. II've identified the error in my code, which was indeed caused by recent updates in both unsloth and unsloth_zoo that I didn't notice. So now my notebook should...
> @GAD-cell do you mind pushing the additional changes you made to the same 2752 PR? [unslothai/unsloth#2752](https://github.com/unslothai/unsloth/pull/2752), that way we have the updated modifications in one consolidated file, and make...
> Hello, > > The ASSERT DEVICE CUDA runtime error still showing up on colab-T4 and colab-A100 during GRPO training  > >...
@rolandtannous ok yes my bad for --no-deps. I re-ran the notebook with the following installation : first run extra colab install then install my forks with --no-deps. This worked for...
Yes I didn't implement GRPO with vLLM, but I can try to integrate it. However since TRL released VLM GRPO officially I think you should refer to the official implementation...