bobbai00 opened a new pull request, #4473: URL: https://github.com/apache/texera/pull/4473
### What changes were proposed in this PR? Pin torch to the CPU-only wheel on Linux x86_64 so the proprietary NVIDIA CUDA shared libraries (`nvidia-*-cu12`) are not pulled into the release artifact. PyPI's `torch==2.8.0` wheel for Linux x86_64 transitively installs fourteen `nvidia-*-cu12` packages. Twelve ship under the NVIDIA Software License Agreement (CUDA Toolkit), which is proprietary and not ASF-compatible. Other platforms (macOS, Windows, Linux aarch64) already get a CPU build from PyPI, so this change only affects Linux x86_64. The fix uses PEP 508 markers so resolution differs per platform: ``` --extra-index-url https://download.pytorch.org/whl/cpu torch==2.8.0+cpu ; platform_system == "Linux" and platform_machine == "x86_64" torch==2.8.0 ; platform_system != "Linux" or platform_machine != "x86_64" ``` GPU users install the CUDA wheel themselves post-install. ### Any related issues, documentation, discussions? Closes #4472. ### How was this PR tested? `pip install --dry-run -r amber/operator-requirements.txt` on macOS resolves to plain `torch==2.8.0` with no `nvidia-*` or `triton` packages. On Linux x86_64 the same command resolves to `torch==2.8.0+cpu` from the PyTorch CPU index, also without `nvidia-*` or `triton`. ### Was this PR authored or co-authored using generative AI tooling? Generated-by: Claude Code (Claude Opus 4.7) -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
