ValueError: invalid literal for int() with base 10: '90a'

[MatthewCroughan]

>>> import torch
>>> print(torch.cuda.get_arch_list())
['sm_60', 'sm_61', 'sm_70', 'sm_75', 'sm_80', 'sm_86', 'sm_89', 'sm_90', 'sm_90a']

  File "/home/matthew/git/ai-toolkit/toolkit/optimizers/optimizer_utils.py", line 4, in <module>
    from optimum.quanto import QBytesTensor
  File "/nix/store/ymj1w34dfw2bi3ai9njyxg2q5fr4r18h-python3.12-optimum-quanto-66dca87bc7e61ffbf1a06da9fd324469592e17e9/lib/python3.12/site-packages/optimum/quanto/__init__.py", line 18, in <module>
    from .library import *
  File "/nix/store/ymj1w34dfw2bi3ai9njyxg2q5fr4r18h-python3.12-optimum-quanto-66dca87bc7e61ffbf1a06da9fd324469592e17e9/lib/python3.12/site-packages/optimum/quanto/library/__init__.py", line 15, in <module>
    from .extensions import *
  File "/nix/store/ymj1w34dfw2bi3ai9njyxg2q5fr4r18h-python3.12-optimum-quanto-66dca87bc7e61ffbf1a06da9fd324469592e17e9/lib/python3.12/site-packages/optimum/quanto/library/extensions/__init__.py", line 23, in <module>
    from .cuda import *
  File "/nix/store/ymj1w34dfw2bi3ai9njyxg2q5fr4r18h-python3.12-optimum-quanto-66dca87bc7e61ffbf1a06da9fd324469592e17e9/lib/python3.12/site-packages/optimum/quanto/library/extensions/cuda/__init__.py", line 54, in <module>
    quanto_cuda_arch = get_max_cuda_arch()
                       ^^^^^^^^^^^^^^^^^^^
  File "/nix/store/ymj1w34dfw2bi3ai9njyxg2q5fr4r18h-python3.12-optimum-quanto-66dca87bc7e61ffbf1a06da9fd324469592e17e9/lib/python3.12/site-packages/optimum/quanto/library/extensions/cuda/__init__.py", line 31, in get_max_cuda_arch
    supported_sm = [int(arch.split("_")[1]) for arch in torch.cuda.get_arch_list() if "sm_" in arch]
                    ^^^^^^^^^^^^^^^^^^^^^^^
ValueError: invalid literal for int() with base 10: '90a'

[dacorvo]

Yes, the code for parsing the arch list does not expect the a. It should be relatively easy to fix as a workaround, even if a more robust algorithm to decide if we can proceed with compilation is definitely required.