build: bump transformer-engine to release_v2.16_post#5507
Conversation
Pin transformer-engine to release_v2.16_post (d64bc14dc87eb658ab98839e4b7687595ee53e2d) in [tool.uv.sources] and regenerate uv.lock. Signed-off-by: oliver könig <okoenig@nvidia.com>
|
This PR has been automatically converted to draft because all PRs must start as drafts. When you are ready for review, click Ready for Review to begin the review process. This will:
See the contribution guide for more details. |
|
Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually. Contributors can view more details about this message here. |
|
/ok to test fc3817e |
|
🔄 Merge queue validation started! You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/28260354930 |
|
🔄 Merge queue validation started! You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/28267788935 |
|
Superseded by #5517, which pins the |
Background
release_v2.16_postto pick up the latest fixes on the 2.16 line.What changed
transformer-enginegit pin in[tool.uv.sources].uv.lock.Details
pyproject.toml:rev4220403e→d64bc14d— the resolved HEAD of TErelease_v2.16_post, pinned as a full SHA for reproducibility.uv.lock: only the TEsource/gitrefs move to the new rev; no other resolution changes (2 files, 8 lines total).Tested
uv lockregenerated cleanly inside the CI dev container (uv 0.7.2); diff limited to the TE rev.Run functional tests+container::lts— a TE bump can shift training numerics, so exercise the golden-value compare on both image variants.Cherry-pick
core_r0.18.0for auto-cherrypick to the release branch after merge.