Possible values of TORCH_CUDA_ARCH_LIST

What are the possible values of this environment variable? And what do they mean? I could not find a documentation page.
I have seen these examples over the internet:
TORCH_CUDA_ARCH_LIST=Turing
TORCH_CUDA_ARCH_LIST=“6.1 8.6”

You can specify GPU architectures and compute capabilities using this env variable to build native PyTorch CUDA kernels for the desired GPUs. Check the Compute Capability section to see which GPU uses which architecture.