[Feature] Request Qwen-Image-Pruning ~

[juntaosun]

Checklist

• 1. If the issue you raised is not a feature but a question, please raise a discussion at https://github.com/mit-han-lab/nunchaku/discussions/new/choose. Otherwise, it will be closed.
• 2. I will do my best to describe the issue in English.

Motivation

Qwen-Image-Pruning is a model made by OPPO after pruning.
The model size is about 30% - 50% smaller than the original, and the effect is almost the same.
Looking forward to uploading the pruning version nunchaku ~

Original model size comparison:
Qwen-Image: 40Gb
Qwen-Image-Pruning: 28Gb

https://huggingface.co/OPPOer/Qwen-Image-Pruning

https://github.com/OPPO-Mente-Lab/Qwen-Image-Pruning

Related resources

No response

[Heliumrich]

Afaik qwn image with nunchaku alrrady run on 6gb vram so...

[frankyifei]

nunchaku needs 15gb vram to run qwen image without offloading to CPU, this would make qwen run much faster as it reduce offloading

[zwukong]

体积大幅缩减，要是能支持一下就完美了，估计能达到flux的大小

[zwukong]

内存终于不用吃紧了，qwen这个模型非常完美，只可惜太大，这个模型绝对是救星

[juntaosun]

@lmxyy I hope that nunchaku can launch a universal and convenient quantification tool (nunchaku conversion tool) so that everyone can quantify the models they need, and better accelerate the improvement and rapid development of the nunchaku ecosystem.

[lmxyy]

will try to do so.

[github-actions]

This issue has been automatically closed due to 30-day inactivity. Please feel free to reopen it with /reopen if needed.