Prebuilt wheels provided via 3rd party repository
#5,528 opened on 2024年7月22日
説明
Prebuilt wheels for PyTorch packages with custom ops
I've created a repository that can build PyTorch wheels with custom ops through the GitHub Actions pipeline and publish them using GitHub Releases, with build attestation. Check it out at https://github.com/MiroPsota/torch_packages_builder.
Since there are various ways how to use it, please refer to the repository README for more information.
If you prefer own build or can't trust a 3rd party repository, feel free to fork it and build any package/version/commit ID you desire yourself.
- Pip cache might not work: pip relies on http cache, and GitHub generates on-the-fly redirections for release links, so they are probably not playing nicely together. If you use uv, you can set custom cache policy.
Fairseq specific info:
Install using pip:
pip install --extra-index-url https://miropsota.github.io/torch_packages_builder fairseq==<version>+<OPTIONAL_commit_hash>pt<PyTorch_version><compute_platform>
Where <compute_platform> is, as in PyTorch, one of cpu, cu<CUDA_short_version> (e.g. cu130, cu128, cu118), or rocm<ROCM_version> (not supported right now).
For example, the newest fairseq commit (as of writing) 3d262bb, PyTorch 2.10.0 with CUDA 13.0:
pip install --extra-index-url https://miropsota.github.io/torch_packages_builder fairseq==0.12.2+3d262bbpt2.10.0cu130
Look at releases section if there are any other combinations, I will probably build occasionally with new pytorch releases and versions/commits.
These wheels are built with PyTorch versions 2.3.0 to 2.10.0 (check my GitHub releases for older versions) and their respective compute platforms and supported operating systems. Please note an exception for the ROCm platform. The build is done using the available public GitHub runners, so they might not work on older OSes.
Although the wheels have been successfully built, I don't test them to work correctly.
If you've installed PyTorch with pip, there's no need to have CUDA installed on your system, as the PyTorch wheels for pip bundle CUDA.