verl Mulit-modal rl training support?

Will support with multimodal training in rl?

Jan 30 '25 14:01 lucasjinreal

Could you provide a list of models you would like us to support? Please also rank them. Thanks.

Jan 30 '25 15:01 vermouth1992

Hi, I think support LLaVA could be enough.

Jan 31 '25 03:01 lucasjinreal

Qwen vl 2 and 2.5 would be great 👍

Feb 03 '25 09:02 Benjoyo

It is great to support Qwen2-VL and Qwen2.5-VL, which are two popular series of models when exploring "aha moment" in MLLM. Thank you for your great work.

Feb 10 '25 16:02 tzjtatata

qwen 2/2.5 vl please. Thanks.

Feb 23 '25 10:02 Adenialzz

qwen vl series is being worked on by a member from the community. the example runs fine and will be soon open sourced

Feb 23 '25 23:02 eric-haibin-lin

Can InternVL2.5 be added to the supported models?

Feb 24 '25 12:02 miaodog

希望LLaVA也被支持，谢谢

Mar 05 '25 08:03 ymxyll

qwen 2.5 vl is supported and available in https://github.com/volcengine/verl/blob/main/examples/grpo_trainer/run_qwen2_5_vl-7b.sh Contribution from the community is highly appreciated

Apr 06 '25 19:04 eric-haibin-lin

Get it! Thank you for your great works!

---Original--- From: @.> Date: Mon, Apr 7, 2025 03:11 AM To: @.>; Cc: "Yuanze @.@.>; Subject: Re: [volcengine/verl] Mulit-modal rl training support? (Issue #168)

qwen 2.5 vl is supported and available in https://github.com/volcengine/verl/blob/main/examples/grpo_trainer/run_qwen2_5_vl-7b.sh Contribution from the community is highly appreciated

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***> eric-haibin-lin left a comment (volcengine/verl#168)

qwen 2.5 vl is supported and available in https://github.com/volcengine/verl/blob/main/examples/grpo_trainer/run_qwen2_5_vl-7b.sh Contribution from the community is highly appreciated

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

Apr 07 '25 02:04 tzjtatata

Hi, I think support LLaVA could be enough.

Llava grpo is supported and available in https://github.com/PRIS-CV/GRPO-for-Llava ❤️

Apr 07 '25 14:04 LeonDiao0427

Could you also support https://huggingface.co/Qwen/Qwen2.5-Omni-7B and https://huggingface.co/Qwen/Qwen2-Audio-7B ?

Thank you very much! Audio modality is also very helpful.

Apr 28 '25 03:04 chunhuizng

mark

Aug 21 '25 02:08 wulaoshi

Could you also support https://huggingface.co/Qwen/Qwen2.5-Omni-7B and https://huggingface.co/Qwen/Qwen2-Audio-7B ?

Omni support is on the way :) https://github.com/volcengine/verl/issues/2852

Sep 01 '25 12:09 TomQunChao

Would also be great to see the new Qwen3-Omni models: https://huggingface.co/collections/Qwen/qwen3-omni-68d100a86cd0906843ceccbe. This may be complicated by those being MoE

Sep 24 '25 23:09 dannnnthemannnn

希望llava-video-qwen2也可以被支持

Jan 12 '26 18:01 SH9959