[BE] Remove the default TORCH_CUDA_ARCH_LIST in CI Docker image (#161137)

This doesn't make sense to have this default to Maxwell, which is too old.  All other places in CI/CD needs to overwrite this value.  IMO, it makes more sense to not set this at all and let CI/CD jobs set it for their own use cases instead.  This is partly responsible for the build failure in https://github.com/pytorch/pytorch/issues/160988
Pull Request resolved: https://github.com/pytorch/pytorch/pull/161137
Approved by: https://github.com/msaroufim

.ci/docker/ubuntu/Dockerfile

@@ -181,7 +181,6 @@ COPY --from=pytorch/llvm:9.0.1 /opt/llvm /opt/llvm
 RUN if [ -n "${SKIP_LLVM_SRC_BUILD_INSTALL}" ]; then set -eu; rm -rf /opt/llvm; fi
 
 # AWS specific CUDA build guidance
-ENV TORCH_CUDA_ARCH_LIST Maxwell
 ENV TORCH_NVCC_FLAGS "-Xfatbin -compress-all"
 ENV CUDA_PATH /usr/local/cuda
 

.ci/pytorch/common_utils.sh

@@ -152,8 +152,6 @@ function get_pinned_commit() {
 function install_torchaudio() {
   local commit
   commit=$(get_pinned_commit audio)
-  # TODO (huydhn): PyTorch CI docker image set the default TORCH_CUDA_ARCH_LIST
-  # to Maxwell. This default doesn't make sense anymore and should be cleaned up
   if [[ "${BUILD_ENVIRONMENT}" == *cuda* ]] && command -v nvidia-smi; then
     TORCH_CUDA_ARCH_LIST=$(nvidia-smi --query-gpu=compute_cap --format=csv | tail -n 1)
     export TORCH_CUDA_ARCH_LIST