Qwen/Qwen-Image · Update README.md to include a quantization code snippet

about 9 hours ago

•

sayakpaul changed pull request title from Update README.md to Update README.md to include a quantization code snippet about 9 hours ago

about 6 hours ago

•

I tested this PR with the latest diffusers, bitsandbytes, and transformers from git and I think there might be some issues.

Steps is 4 in the quantized example which I believe is wrong. The non-quantized example uses 50 steps and using 4 steps with this code just results in a blur.
This model seems extremely sensitive to NF4 quantization, and running this code (even modified for 50 steps) results in a very grainy image.
The PR'd example removes the Ultra HD, 4K, cinematic composition. suffix which seems to be needed to get decent (if grainy) results.

EDIT: for anyone looking to run this quantized, Optimum Quanto (code taken from example here https://github.com/QwenLM/Qwen-Image/pull/6/files) seems to work much better than bitsandbytes NF4.