Does this support Lightning LoRAs?
Hello, I just want to ask does this quantization support lightx2v/Qwen-Image-Lightning? If yes, should I just load it in using diffusers' function or is there a specific way to load this in? I'm not really used to working with quantization version so sorry if this question may sound unintuitive.
Loras should work but I don’t think this particular speed up Lora works. I had given it a try.
Since 4 bits is already much faster and smaller I saw no point in trying to add speed up Loras.
Thank you for quick response. I just want to see if we can speed it up a little bit more. Well it's an effort anw. Can I ask what script do you use to quantize this model and do you mind sharing it? Qwen just release a new version so I want to give it a try.
its not a particular script but Its something that is made specifically to target a model through multiple tests. What hardware resources do you have? You need atleast 1 H100 or similar gpu to do the quantization.
you can check the documentation of BNB and build your own quantization.
I did t plan to work during holidays, but I have published it for 2511.