Why is the FP8 unet file the same size as the BF16 unet file?
#1
by
Nap
- opened
I checked the SHA256 and they are indeed different files, why is are they both 23.8 GB?
i think its mistake !
The fp8 model uses bf16 weights but adds fp8 Q/DQ ops, so the size of the ONNX model won't be smaller on disk (the trt engine, however, will be smaller)
timudk
changed discussion status to
closed
Great, thanks for the info.