ci: bump Primus-Turbo/AITER pins for Flux diffusion (mxfp4) + skip CI on draft PRs#806
Merged
Conversation
… on draft PRs Bump PRIMUS_TURBO_COMMIT to Primus-Turbo main (0.3.x, validated build) which exposes the mxfp4 gemm_fp4_impl(..., preshuffled=...) fast path, and PRIMUS_TURBO_AITER_COMMIT to AITER v0.1.14.post1. Mirror the turbo commit into benchmark.yaml; the docker build receives both via ci.yaml build-args. Add a draft-skip guard to the expensive jobs (build-docker, run-unittest-torch, run-unittest-jax) so the stacked Flux PRs can sit in Draft without firing the full pipeline; code-lint stays ungated. Add ready_for_review to the pull_request trigger types so flipping a draft to Ready triggers the first run. MEGATRON_PATH export intentionally omitted: the Flux diffusion unit tests run in-process (no megatron-importing subprocess remains after the te_vs_local in-process conversion), and tests/conftest.py auto-adds third_party/Megatron-LM to sys.path when MEGATRON_PATH is unset. CI already checks out submodules: recursive, so megatron imports without any extra env.
jasainio
requested review from
Xiaoming-AMD,
limou102 and
wenxie-amd
as code owners
… DeepEP pad_multiple
The branch's pins pointed at older commits missing required APIs, breaking
DeepEP MoE training with:
DeepEPTokenDispatcher.__init__() got an unexpected keyword argument 'pad_multiple'
- PRIMUS_TURBO_COMMIT 3cd482de -> bc6c7d2a :
old pin lacked both gemm_fp4_impl(preshuffled=) and
DeepEPTokenDispatcher(pad_multiple=); new pin has both.
- PRIMUS_TURBO_AITER_COMMIT f299f579 -> 0f3c58e6: the real v0.1.14.post1
tag commit required by Primus-Turbo main's aiter_utils.py.
Contributor
|
Update: Primus-Turbo/AITER pins were aligned with main branch and tests pass |
Xiaoming-AMD
approved these changes
Jul 6, 2026
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
5 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.
Root of an 18-PR series splitting the Flux diffusion-training feature (training Flux, a DiT text-to-image diffusion model, on Primus/Megatron) out of one large branch for reviewability. This PR targets
mainand is the first to merge. The content branches are cut from this branch (feat/flux/ci-env), not rawmain, so they carry the bumped pins + draft guard on their own heads — which lets the rest of the stack open as drafts immediately, before this PR merges.What this changes
Updates the public-CI dependency pins to the versions the Flux diffusion feature needs:
PRIMUS_TURBO_COMMITto a recent Primus-Turbomainbuild that exposes the mxfp4gemm_fp4_impl(..., preshuffled=...)fast path, andPRIMUS_TURBO_AITER_COMMITto AITERv0.1.14.post1. Mirrors the same turbo commit intobenchmark.yaml(the docker build picks both up throughci.yaml's build-args).build-docker,run-unittest-torch,run-unittest-jax) so the rest of the stack can be opened as Draft PRs without firing the full docker-build + GPU pipeline on every open/sync;code-lintstays ungated. Two parts: addready_for_reviewto thepull_requesttriggertypes(the defaultopened/synchronize/reopenedset omits it, so without this the draft→ready flip triggers nothing) and addif: ${{ github.event_name != 'pull_request' || !github.event.pull_request.draft }}to the three heavy jobs.Why it merges first (but need not block opening)
On the current
mainpins the mxfp4 path fails (gemm_fp4_implrejects thepreshuffledarg), so content PRs can only go green if their head branch carries the new pins — which stacking onfeat/flux/ci-envprovides without a merge. Merging this first matters at merge time: once it's onmain, children auto-retargeted tomaininherit the fix and the guard.No
MEGATRON_PATHchange is needed: the diffusion unit tests run in-process andtests/conftest.pyalready puts thesubmodules: recursive-checked-outthird_party/Megatron-LMonsys.pathwhenMEGATRON_PATHis unset.Dependencies
None — this is the root prerequisite.
Test plan
No unit tests of its own. After the runner rebuilds the image, confirm a smoke run is green:
pytest tests/unit_tests/backends/megatron/diffusion -k mxfp4. The guard can be sanity-checked by opening a throwaway draft PR (expect onlycode-lint) and marking it ready (expect the full pipeline).Files
2 (
.github/workflows/ci.yaml— pins + draft guard;.github/workflows/benchmark.yaml— mirrored turbo pin).