Benchmark vs Q8 GGUF?
#1
by Adreitz - opened
This is the only ConvRot quant of Flux.2 Dev that I've found that works natively in ComfyUI. Thanks for making it! Since you benchmarked the performance against bf16 and compared the PSNR to fp8 and nvfp4, would you be willing to check the PSNR of a Q8 GGUF using your same procedure? I'd be interested to see the accuracy differences, especially considering that your quant is somehow about 2GB smaller than the GGUF I have been using, https://huggingface.co/city96/FLUX.2-dev-gguf.
Edit: There is also this one by Unsloth, which I would tend to trust more, but I haven't tried it: https://huggingface.co/unsloth/FLUX.2-dev-GGUF. Nevermind, the two Q8 files have the exact same hash, so they are the same file.