Build sdist and wheel in CI - #1865
Open
calebho wants to merge 12 commits into
Open
Conversation
calebho
marked this pull request as ready for review
December 17, 2024 05:03
Author
|
cc @crcrpar |
Contributor
|
I think this repo is a better implementation to build apex wheels. |
bvolpato
reviewed
Aug 16, 2026
bvolpato
left a comment
bvolpato
left a comment
There was a problem hiding this comment.
Could we adjust Python/container matrix before enabling this? I reproduced CUDA 12.4 Ubuntu 20.04 job and deadsnakes only provided 3.9 dev/venv packages. Python 3.10, 3.11, and 3.12 fail at apt install, leaving 9 of 12 combinations unable to reach build.
Could MANIFEST.in also include apex/contrib/csrc? Generated sdist omitted those CUDA sources, and building --fast_layer_norm from extracted archive failed on missing ln_api.cpp.
Is intended scope CI artifacts only? Current workflow uploads unsigned x86_64 artifacts, but does not publish release wheels or cover ARM/CUDA 13.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
summary
People often run into issues building from source, so it would help if there were pre-built wheels with sensible defaults. This PR adds GitHub Actions that builds wheels for various Python and CUDA versions and saves them to the workflow artifacts on each commit push.
To get the wheel, one needs to go to the workflow page and download it following https://docs.github.com/en/actions/managing-workflow-runs-and-deployments/managing-workflow-runs/downloading-workflow-artifacts. For example,
Still clunky, but much faster than building from source
The specific versions were chosen based on what is currently supported in PyTorch stable (2.5 at the time of writing). Specifically:
Notes
Good follow ups
pip install https://github.com/NVIDIA/apex/releases/...on the appropriate wheeltest plan
builds are green: https://github.com/calebho/apex/actions/runs/12365956032/job/34511835345
ran L0 tests on Python 3.10 + CUDA 12.1 on 2xA100 40GB