Help running on CPU

#1
by liambarryarm - opened

I want to run models via Amphion on CPU only. I could not find a way to specify this during the setup and was forced to download several python packages related to GPU to execute run.sh. Unfortunately it fails immediately with:

ImportError: python3.10/site-packages/torch/lib/libtorch_cuda.so: undefined symbol: ncclCommRegister

I should not need this package to run on CPU - are there any plans to support CPU or can you comment on whether it is too slow to be of practical use?

Liam Barry (Arm Employee)

Sign up or log in to comment