Summary
Installing mamba-ssm can fail during wheel build when setup.py:get_wheel_url() parses vendor-specific PyTorch/CUDA version metadata with packaging.version.parse. If the environment contains a non-PEP-440 vendor build string, get_wheel_url() raises packaging.version.InvalidVersion and aborts the install instead of falling back to a source build.
This is reproducible in vendor PyTorch containers where CUDA/PyTorch build metadata may be exported as values like gpgpu.<build-id>.
Observed failure
Package:
mamba-ssm==2.3.2.post1
- Python 3.12
- Linux x86_64
- Vendor/internal PyTorch prerelease container
During installation as a transitive dependency:
Building mamba-ssm==2.3.2.post1
× Failed to build `mamba-ssm==2.3.2.post1`
...
File "<string>", line 340, in run
File "<string>", line 315, in get_wheel_url
File ".../packaging/version.py", line 103, in parse
return Version(version)
File ".../packaging/version.py", line 361, in __init__
raise InvalidVersion(f"Invalid version: {version!r}")
packaging.version.InvalidVersion: Invalid version: 'gpgpu.<build-id>'
Summary
Installing
mamba-ssmcan fail during wheel build whensetup.py:get_wheel_url()parses vendor-specific PyTorch/CUDA version metadata withpackaging.version.parse. If the environment contains a non-PEP-440 vendor build string,get_wheel_url()raisespackaging.version.InvalidVersionand aborts the install instead of falling back to a source build.This is reproducible in vendor PyTorch containers where CUDA/PyTorch build metadata may be exported as values like
gpgpu.<build-id>.Observed failure
Package:
mamba-ssm==2.3.2.post1During installation as a transitive dependency: