Skip to content

INFERENG-10518 (fix): Align DeepEP wheel version with rhaiv.4 - #6

Merged
mwoodson merged 1 commit into
neuralmagic:mainfrom
mwoodson:mwoodson/infereng-10518-deepep-rhaiv4
Sep 25, 2026
Merged

mwoodson merged 1 commit into
neuralmagic:mainfrom
mwoodson:mwoodson/infereng-10518-deepep-rhaiv4

Conversation

@mwoodson

@mwoodson mwoodson commented Sep 25, 2026 •

Copy link
Copy Markdown

TL;DR

[PROPOSAL] Align DeepEP's midstream wheel-version helper and static project metadata on 2.0.1+rhaiv.4. The existing annotated .3 source tag contains rhai_version = "0", while Fast-2 requires an exact local-version wheel; this draft must not be tagged or consumed until an actual CUDA-built wheel reports the matching METADATA Version.

  • Keep the runtime public deep_ep.__version__ at 2.0.1; this PR changes packaging metadata only.
  • Do not move or reuse v2.0.1+rhaiv.3. A new tag is a separate release-owner action after review and wheel verification.

Changes

File Change
setup.py Bump the midstream local suffix from 0 to 4.
pyproject.toml Align the static project version to 2.0.1+rhaiv.4.

Jira

INFERENG-10518: Fast-2 CUDA release validation. This packaging mismatch blocks binary-only installation on x86_64 and aarch64.

Test Plan

  • Passed locally: git diff --check, python -m py_compile setup.py, TOML parsing, and an isolated call to get_package_version() confirming it agrees with [project].version at 2.0.1+rhaiv.4.
  • Required before making this PR ready or tagging: build the wheel in the CUDA-enabled DeepEP/Fondue build environment and inspect its filename and .dist-info/METADATA for Version: 2.0.1+rhaiv.4. The repository formatting check does not establish wheel metadata.
  • After maintainer review/merge and a new immutable tag: verify GitLab mirror SHA parity, then update the RHAIIS pipeline pin and confirm both architecture wheel releases install with their generated requirements and constraints. No tag or downstream release is part of this PR.

Keep the midstream local suffix and static project metadata consistent so a release built from the next tag can satisfy the exact downstream DeepEP requirement. The final wheel filename and METADATA version still require CUDA-builder verification before tagging.

Signed-off-by: Matt Woodson <mwoodson@redhat.com>
@mwoodson
mwoodson marked this pull request as ready for review September 25, 2026 13:47
@mwoodson
mwoodson merged commit 578f2f5 into neuralmagic:main Sep 25, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants