ci: bump base image/runner to v26.4 and drop v26.3-only workarounds#896
Draft
WangLingxun wants to merge 2 commits into
Draft
ci: bump base image/runner to v26.4 and drop v26.3-only workarounds#896WangLingxun wants to merge 2 commits into
WangLingxun wants to merge 2 commits into
Conversation
WangLingxun
requested review from
Xiaoming-AMD,
limou102 and
wenxie-amd
as code owners
July 21, 2026 05:57
WangLingxun
marked this pull request as draft
July 21, 2026 06:21
- ci.yaml: BASE_IMAGE + runs-on v26.3 -> v26.4; drop the v26.3-only 'Purge stale TorchTitan' step (v26.4 ships no torchtitan to shadow). Keep 'Install fixed origami' (v26.4 still ships no origami). - Dockerfile: default BASE_IMAGE + comments v26.3 -> v26.4. - docs/scripts: bump default image refs to v26.4, release branch to release/v26.4, and pip pin to primus==26.4.0.
WangLingxun
force-pushed
the
chore/ci-bump-primus-v26.4
branch
from
July 21, 2026 06:46
5b61b49 to
c6d16f3
Compare
- Dockerfile/ci.yaml: drop the origami (rocm-libraries@223648a) install. This base image already ships no origami, so primus_turbo falls back to heuristic kernel selection for MoE grouped-gemm (benchmarked: no measurable gain on gfx942). Comments show the pip install one-liner to add it back if ever needed (e.g. gfx950 tuning). - aiter_deepbind_patches.py: _can_install_aiter_deepbind now also checks transformer_engine.__version__ and skips installing the RTLD_DEEPBIND hook once TE >= 2.14 (confirmed fixed in rocm/primus:v26.4: vendored lib renamed, no longer exports the colliding aiter::mha_bwd symbol). Unknown/unparseable versions fail safe towards 'still install it'. This keeps the patch a no-op on new images while remaining the required fix for anyone still on rocm/primus:v26.3 or older, without needing separate per-arch hardware validation (same TE wheel serves gfx942/gfx950). - Add unit tests for the new gating logic (no GPU required).
WangLingxun
force-pushed
the
chore/ci-bump-primus-v26.4
branch
from
July 21, 2026 09:24
a6e9036 to
90e0389
Compare
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
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.
Upgrade the Primus runtime image from rocm/primus:v26.3 to v26.4 across CI Dockerfile/workflow, runner config, example launch scripts, benchmark scripts and docs (14 files). JAX/MaxText base image not upgraded this round.
Image bump
Workaround cleanup
aiter_deepbind_patches.pynow self-detects whether it's still needed:_can_install_aiter_deepbindalso checkstransformer_engine.__version__and skips installing the RTLD_DEEPBIND hook once TE >= 2.14 (confirmed fixed in v26.4 — the vendored lib was renamed and no longer exports the collidingaiter::mha_bwdsymbol). Unknown/unparseable versions fail safe towards "still install it". This keeps the patch a no-op on v26.4+ while remaining the required fix for anyone still on rocm/primus:v26.3 or older, without needing separate per-arch hardware validation (the same TE wheel serves gfx942/gfx950). Added unit tests for the new gating logic.