infra: add manual release-wheel builder (Hopper); no tag-triggered CI - #323
Draft
Andrewxu313 wants to merge 1 commit into
Draft
Andrewxu313 wants to merge 1 commit into
Andrewxu313 wants to merge 1 commit into
Conversation
Builds the complete Hopper reference wheel set (flash-attn 3, FlashMLA, DeepGEMM, batchgen_kernels sm90a, batchgen) via the same pinned recipes as install_deps.sh but emitting wheels, and optionally uploads them to a GitHub release tag with --tag. Intended to be run manually on a Hopper node (e.g. h200-instance-3) before a release, since there is no stable GPU CI runner. Pairs with the install_deps.sh wheel fast path.
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Description
Add
scripts/build_release_wheels.sh— a manual release-wheel builder run on a Hopper node (e.g. h200-instance-3) before a release. It builds the complete reference wheel set — flash-attn 3, FlashMLA, DeepGEMM, batchgen_kernels (sm90a), batchgen — using the same pinned recipes asinstall_deps.shbut emitting wheels (pip wheel --no-deps), and optionally uploads them all to one GitHub release tag (--tag, viagh). This is the setinstall_deps.sh's auto-download fast path (#322) looks for on a single tag.No tag-triggered GitHub Actions workflow is added: there is no stable GPU runner, so per the current decision the wheel build is run by hand before each release.
Motivation
The canonical distribution is source install; the wheel fast path (#322) only activates when the complete wheel set lives on one release tag. Today those wheels are hand-built and split across tags. This script makes producing the full, co-located set a single repeatable command on a Hopper box.
bash -n+--help. Needs one run on h200-instance-3 to confirm the builds (esp. the FA3 Hopper build) and the resulting wheel tags. Re-hosting FA3 (Dao-AILab, BSD-3) / FlashMLA / DeepGEMM (DeepSeek, MIT) wheels is redistribution of permissively-licensed deps (same as INSTALL.md Option C already does).Type of Change
model— add/extend model support (models/**+ registration seam + model kernels only)kernel— add/optimize a compute kernel (batchgen_kernels/**+ in-tree kernel dirs)core— change scheduling/serving/runtime scaffolding (the only type that may)fix— narrow bug fix (+ a regression test)infra— build / CI / packaging / scripts / Dockerdocs— documentation onlyFile changes
scripts/build_release_wheels.sh--tagupload); mirrors install_deps.sh pinsChecklist