The "Install CPU-only torch" step fails on every run.
Cause
Debian bookworm's python3-venv seeds the venv with pip 23.0.1. That pip compares the requested name typing-extensions with the wheel's metadata name typing_extensions without normalising them, so it discards the wheel the PyTorch CPU index serves as "inconsistent Name". It then falls back to the source distribution, whose build needs flit_core, and the PyTorch-only index doesn't carry it. The step fails.
Fix
Upgrade pip in the venv, from PyPI, before the torch install.
Testing
Reproduced locally with the same pip 23.0.1: it fails with the same flit_core error.
Current pip (26.2.1) downloads the typing_extensions wheel from the same index without complaint.
Not verified: a full resolve of torch torchaudio. My sandbox can't reach download-r2.pytorch.org, where the wheels themselves are hosted.
Under pull_request_target the workflow runs from main, so this PR's own run still uses the old workflow and will fail at the same step. Once this is merged, re-running #24's checks will use the fix.
**Summary generated by LLM agent**
The "Install CPU-only torch" step fails on every run.
## Cause
Debian bookworm's `python3-venv` seeds the venv with pip 23.0.1. That pip compares the requested name `typing-extensions` with the wheel's metadata name `typing_extensions` without normalising them, so it discards the wheel the PyTorch CPU index serves as "inconsistent Name". It then falls back to the source distribution, whose build needs `flit_core`, and the PyTorch-only index doesn't carry it. The step fails.
## Fix
Upgrade pip in the venv, from PyPI, before the torch install.
## Testing
- Reproduced locally with the same pip 23.0.1: it fails with the same `flit_core` error.
- Current pip (26.2.1) downloads the `typing_extensions` wheel from the same index without complaint.
- Not verified: a full resolve of `torch torchaudio`. My sandbox can't reach `download-r2.pytorch.org`, where the wheels themselves are hosted.
- Under `pull_request_target` the workflow runs from `main`, so this PR's own run still uses the old workflow and will fail at the same step. Once this is merged, re-running #24's checks will use the fix.
Debian bookworm's python3-venv seeds pip 23.0.1, which rejects the
PyTorch CPU index's typing_extensions wheel because its metadata spells
the name with an underscore. It then falls back to the sdist, whose build
needs flit_core, which that index does not carry.
branderson
approved these changes 2026-10-10 21:50:55 -07:00
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Summary generated by LLM agent
The "Install CPU-only torch" step fails on every run.
Cause
Debian bookworm's
python3-venvseeds the venv with pip 23.0.1. That pip compares the requested nametyping-extensionswith the wheel's metadata nametyping_extensionswithout normalising them, so it discards the wheel the PyTorch CPU index serves as "inconsistent Name". It then falls back to the source distribution, whose build needsflit_core, and the PyTorch-only index doesn't carry it. The step fails.Fix
Upgrade pip in the venv, from PyPI, before the torch install.
Testing
flit_coreerror.typing_extensionswheel from the same index without complaint.torch torchaudio. My sandbox can't reachdownload-r2.pytorch.org, where the wheels themselves are hosted.pull_request_targetthe workflow runs frommain, so this PR's own run still uses the old workflow and will fail at the same step. Once this is merged, re-running #24's checks will use the fix.