Hey, great work you have been doing. I've been following along for a while now and just got around to converting models using your methods.
I was wondering if you have done any of the V2 scaled versions for sdxl or if its just a future scope idea? I am familiar with the scaler layers from NVFP4 quantization so I am thinking of moving towards giving it a go.
I also gave your methods a go for sd15 and it went well so far, https://huggingface.co/ApacheOne/HSWQ-fp8-SD15. Modern comfyUI gives bad outputs for sd15 so Im in the process of setting up old A1111 for better testing.
Thank you.
Hey, great work you have been doing. I've been following along for a while now and just got around to converting models using your methods.
I was wondering if you have done any of the V2 scaled versions for sdxl or if its just a future scope idea? I am familiar with the scaler layers from NVFP4 quantization so I am thinking of moving towards giving it a go.
I also gave your methods a go for sd15 and it went well so far, https://huggingface.co/ApacheOne/HSWQ-fp8-SD15. Modern comfyUI gives bad outputs for sd15 so Im in the process of setting up old A1111 for better testing.
Thank you.