Installation Error

[dannyxiaocn]

When I install vLLM, I have pull the docker as the tutorial and install from source. But there are an error encountered, how can I fix this:
ValueError: Unsupported CUDA arch (5.2). Valid CUDA arch strings are: ['7.0', '7.5', '8.0', '8.6', '8.9', '9.0', '7.0+PTX', '7.5+PTX', '8.0+PTX', '8.6+PTX', '8.9+PTX', '9.0+PTX'].

[MarcCote]

I managed to get it working by setting ENV TORCH_CUDA_ARCH_LIST="8.0;8.6;8.9" (depends on your GPU) but I think an ENV TORCH_CUDA_ARCH_LIST="" could also work.

I believe the check here should be a

if none of the architectures in arch_list are in valid_arch_strs, then raise an error

instead of

if one of the architectures in arch_list is not in valid_arch_strs, then raise an error