new model support

[simplew2011]

• [ ] DeepseekV3
• [x] Qwen3-30B-A3B
• [ ] Qwen3-Next-80B-A3B-Instruct
• [x] Qwen3_VL

[wangzhaode]

Thanks for reaching out and suggesting these new models. Here is the current status:

• DeepseekV3: This model is too large for on-device deployment, so we will not be supporting it for the time being.
• Qwen3-30B-A3B: This model is already supported and you can use it directly.
• Qwen3-Next-80B-A3B-Instruct & Qwen3_VL: Support for these models is planned. We will prioritize their integration as soon as smaller, official parameter versions are released.

We will keep this issue updated with our progress.

[altunenes]

Thanks for reaching out and suggesting these new models. Here is the current status:

• DeepseekV3: This model is too large for on-device deployment, so we will not be supporting it for the time being.
• Qwen3-30B-A3B: This model is already supported and you can use it directly.
• Qwen3-Next-80B-A3B-Instruct & Qwen3_VL: Support for these models is planned. We will prioritize their integration as soon as smaller, official parameter versions are released.

We will keep this issue updated with our progress.

That would be great! Especially for the smaller VL models. This week, Qwen released some very cool small VL models. I tried converting them myself, but I couldn't manage it. they have some custom components which I'm not aware of tbh. DeepStack, Interleaved-MRoPE etc, dunno if they have ONNX equivalents... :(

[wangzhaode]

Qwen3-VL has been supported on v0.0.4.

MNN Models:

https://huggingface.co/collections/taobao-mnn/qwen3-vl-mnn https://modelscope.cn/collections/Qwen3-VL-MNN-f4da0cedb82847

[altunenes]

amazing! thank you :-)