fp4 transfomer only?

#12
by shadovv76 - opened

Could you extract fp4 transformer only to safetensors file?

FP4 PLS!❤️

Owner

Added.

Kijai changed discussion status to closed

Added.

Was this done with the nvidia toolchain so users with blackwell hardware could expect hardware acceleration? @Kijai

I have no real insight into this at all, but I asked gpt which seemed to have some "knowledge" about it and states that fp4 vs nvfp4 are not the same thing inherently.
https://chatgpt.com/share/696822c1-2c08-8005-8fa8-c487aab363a2

Owner

It's just extracted from the original fp4 checkpoint, which was in nvfp4, this retains the same quantization.

Sign up or log in to comment