Skip to content

Add automated vLLM and SGLang upstream synchronization - #423

Open
Lanoxia wants to merge 12 commits into
ovg-project:mainfrom
Lanoxia:zixuan/engine-upstream-sync
Open

Add automated vLLM and SGLang upstream synchronization#423
Lanoxia wants to merge 12 commits into
ovg-project:mainfrom
Lanoxia:zixuan/engine-upstream-sync

Conversation

@Lanoxia

@Lanoxia Lanoxia commented Aug 4, 2026

Copy link
Copy Markdown
Contributor

Summary

This PR adds repository-owned synchronization for the OVG-maintained vLLM and
SGLang integration branches.

  • Synchronize each integration branch with its official upstream main on a
    daily schedule or manual dispatch.
  • Rebase the small OVG patch stack and update one stable automation branch.
  • Run structural compatibility checks against the engine interfaces patched by
    kvcached before pushing any update.
  • Open or update one compatibility PR per engine instead of creating duplicate
    PRs.
  • Preserve JSON and Markdown reports for conflicts and failed checks so a
    maintainer or repair agent has actionable input.
  • Skip an engine cleanly until its OVG repository is configured.

The PR also updates the SGLang allocator compatibility layer for the current
split allocator package and num_new_pages calling convention detected during
the upstream exercise.

Configuration

Set these repository variables after the OVG engine forks are ready:

OVG_VLLM_REPOSITORY
OVG_VLLM_BASE_BRANCH
OVG_SGLANG_REPOSITORY
OVG_SGLANG_BASE_BRANCH

The base branches default to kvcached-main. Cross-repository writes use the
OVG_SYNC_TOKEN secret; dry runs do not receive that token.

Validation

  • 19 synchronization, compatibility, workflow-contract, and SGLang interface
    tests passed locally.
  • A real dry run against Lanoxia/vllm:kvcached-main rebased cleanly onto the
    latest vllm-project/vllm:main; compatibility scan and three plugin tests
    passed.
  • A real dry run against Lanoxia/sglang:kvcached-main rebased cleanly onto the
    latest sgl-project/sglang:main; compatibility scan and three plugin tests
    passed.
  • Both exercises completed without pushing temporary automation branches.

Deployment boundary

The workflow is ready without private infrastructure. Final deployment only
requires the OVG engine fork locations and a token permitted to update those
repositories and open synchronization PRs.

@gemini-code-assist

Copy link
Copy Markdown
Contributor

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

@RixinLiu RixinLiu self-assigned this Aug 5, 2026
@RixinLiu RixinLiu mentioned this pull request Aug 5, 2026
55 tasks
@Lanoxia
Lanoxia force-pushed the zixuan/engine-upstream-sync branch 2 times, most recently from 29607f8 to d633ff2 Compare August 10, 2026 10:36
@Lanoxia
Lanoxia marked this pull request as ready for review August 10, 2026 10:37
Comment thread tools/sync_engine_upstream.py
@Lanoxia
Lanoxia force-pushed the zixuan/engine-upstream-sync branch from 94dd649 to 25e233d Compare August 13, 2026 07:24
@RixinLiu RixinLiu added gpu-ci and removed gpu-ci labels Aug 18, 2026
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.

4 participants