TensorRT icon indicating copy to clipboard operation
TensorRT copied to clipboard

Does Flux not support int8?

Open algorithmconquer opened this issue 1 year ago • 1 comments

When I use the command " python demo_txt2img_flux.py prompt
--version "flux.1-dev"
--int8
" run the code, the engine file size is the same with float16, how to solve the problem?

algorithmconquer avatar Oct 17 '24 13:10 algorithmconquer

cc @asfiyab-nvidia

yuanyao-nv avatar Oct 18 '24 22:10 yuanyao-nv

INT8 support has not been added to the Flux pipeline yet. Next release will showcase FP8 and BF16 support

asfiyab-nvidia avatar Oct 28 '24 07:10 asfiyab-nvidia