[REQUEST] Please add support for Qwen3VL

[sujitvasanth]

Problem

Please could you add support for QWEN3VL? it has just been released in 8B and 4B versions.
you already have QWEN3 and 2.5VL so presumably may be minimal stretch - I was going to try to build it myself but thought it needs development anyway..

I built a custom architecture for openCUA7B recently... which is working perfectly.. https://huggingface.co/sujitvasanth/OpenCUA-7B-exl2

Solution

qwenv3VL architecture file dont know if it needs new vision tower inference - presumably not

Alternatives

No response

Explanation

forward compatibility

Examples

No response

Additional context

No response

Acknowledgements

• I have looked for similar requests before submitting this one.
• I understand that the developers have lives and my issue will be answered when possible.
• I understand the developers of this program are human, and I will make my requests politely.

[sujitvasanth]

Qwen3-VL’s vision tower is different from Qwen2.5-VL’s, though both are ViT-family encoders.

Qwen2.5-VL: uses a ViT with window attention and an upgraded MRoPE scheme for spatio-temporal handling.
Hugging Face

Qwen3-VL: keeps a ViT backbone but adds DeepStack (multi-level feature fusion) and an interleaved MRoPE layout, plus updated text-timestamp alignment for video—so the encoder+fusion stack isn’t drop-in compatible with 2.5.
Hugging Face

Note: preprocessing is largely consistent (HF shows Qwen3-VL still uses Qwen2VLImageProcessor for images, with a new video processor), but the encoder + projector/fusion (“bridging”) are changed