[CI] Upgrade vLLM to 20250919 (6d8246aa) and fix some broken issue

[Potabk]

What this PR does / why we need it?

1. This pr bump vllm commit to vllm-project/vllm@6d8246a
2. fix upstream changes [Multimodal] Remove legacy multimodal fields in favor of MultiModalFeatureSpec  vllm#24548 abort multi-modal kwargs, make vllm main and v0.10.2 both adaptable
3. fix metadata_builder changes introduced by [Core/DBO][1/N] Add Dual-Batch Overlap mechanism to VLLM vllm#23693
4. fix structured_outputs_config changes introduced by [Chore] Cleanup guided namespace, move to structured outputs config vllm#22772
5. fix moe_config changes introduced by [Kernel] Delegate construction of FusedMoEQuantConfig to FusedMoEMethodBase subclasses vllm#22537

Co-authored-by: MengqingCao cmq0113@163.com
Co-authored-by: Yikun Jiang yikunkero@gmail.com

• vLLM version: v0.10.2
• vLLM main: vllm-project/vllm@c60e613

[github-actions]

👋 Hi! Thank you for contributing to the vLLM Ascend project. The following points will speed up your PR merge:‌‌

• A PR should do only one thing, smaller PRs enable faster reviews.
• Every PR should include unit tests and end-to-end tests ‌to ensure it works and is not broken by other future PRs.
• Write the commit message by fulfilling the PR description to help reviewer and future developers understand.

If CI fails, you can run linting and testing checks locally according Contributing and Testing.

[gemini-code-assist]

Code Review

This pull request appears to fix CI issues by adapting the code to a newer version of vLLM, particularly around multi-modal input handling. The changes introduce version-specific logic to maintain backward compatibility. My review focuses on improving the maintainability of this new logic by reducing code duplication and fixing a potential bug. I've identified two areas where helper functions can be used to create a single, unified implementation for different vLLM versions, which is a pattern already used effectively elsewhere in the changed files.

[codecov]

Codecov Report

❌ Patch coverage is 13.01370% with 127 lines in your changes missing coverage. Please review.
✅ Project coverage is 71.95%. Comparing base (1bbb20e) to head (39eb893).
⚠️ Report is 904 commits behind head on main.

Files with missing lines	Patch %	Lines
vllm_ascend/worker/model_runner_v1.py	3.38%	114 Missing ⚠️
vllm_ascend/worker/npu_input_batch.py	58.33%	5 Missing ⚠️
vllm_ascend/spec_decode/mtp_proposer.py	0.00%	3 Missing ⚠️
tests/ut/ops/test_fused_ops.py	50.00%	1 Missing ⚠️
tests/ut/torchair/ops/test_torchair_fused_moe.py	50.00%	1 Missing ⚠️
vllm_ascend/ops/fused_moe.py	66.66%	1 Missing ⚠️
vllm_ascend/platform.py	66.66%	1 Missing ⚠️
vllm_ascend/torchair/ops/torchair_fused_moe.py	66.66%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2907      +/-   ##
==========================================
- Coverage   74.76%   71.95%   -2.82%     
==========================================
  Files         150      168      +18     
  Lines       20891    23547    +2656     
==========================================
+ Hits        15620    16943    +1323     
- Misses       5271     6604    +1333     

Flag	Coverage Δ
unittests	71.95% <13.01%> (-2.82%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[github-actions]

This pull request has conflicts, please resolve those before we can evaluate the pull request.

[Yikun]

My mean was using the latest hash in here: vllm-project/vllm@68dbde5

Suggested change

	vllm_version: [main, v0.10.2]
	vllm_version: [68dbde5, v0.10.2]

Bump and address upstream change per day,

• pros: this move will shift from being reactive to proactive to avoid community level ci error.
• cons:

1. maintainers should be carefully to review especially some code line
2. we need upgrade main pin hash manually.

[Yikun]

cc @wangxiyuan @MengqingCao @ganyi1996ppo @jianzs @ApsarasX

[Potabk]

ok, by the way, we must use the full commit hash like 68dbde5dbb11b9250454d0c9f21a8b3da960b341, otherwise the checkout@v4 will failed, I have fall into the pit

[Potabk]

Nit: I have an auto workflow to help bump: will submit in this PR or next time, any comments or suggestions are welcome

name: Bump vllm latest commit hash for CI

on:
  schedule:
    - cron: '0 16 * * *'  # At UTC+8 24:00 every day
  workflow_dispatch:

jobs:
  bumper:
    name: Bump vllm latest commit hash for CI
    runs-on: ubuntu-latest
    steps:
      - name: Checkout vllm
        uses: actions/checkout@v4
        with:
          repository: vllm-project/vllm

      - name: Get latest commit hash
        id: get_hash
        run: echo "commit_hash=$(git rev-parse HEAD)" >> $GITHUB_OUTPUT

    outputs:
      commit_hash: ${{ steps.get_hash.outputs.commit_hash }}

  create_pr:
    runs-on: ubuntu-latest
    needs: bumper
    env:
      UPSTREAM_REPO: vllm-project/vllm-ascend
    steps:
      - name: Checkout repository
        uses: actions/checkout@v4
        with:
          repository: vllm-ascend-ci/vllm-ascend
          token: ${{ secrets.PAT_TOKEN }}
          ref: main

      - name: Add upstream remote
        run: |
          git remote add upstream https://github.com/${{ env.UPSTREAM_REPO }}.git
          git fetch upstream
          git remote -v

      - name: Set Git user info dynamically
        run: |
          git config user.name "${{ github.actor }}"
          git config user.email "${{ github.actor }}@users.noreply.github.com"

      - name: Create or switch to branch
        run: |
          TIMESTAMP=$(date +%Y%m%d%H%M%S)
          BRANCH_NAME="auto-pr/Bumper-${TIMESTAMP}"
          echo "BRANCH_NAME=${BRANCH_NAME}" >> $GITHUB_ENV
          git checkout -B "${BRANCH_NAME}" upstream/main
        

      - name: add vllm commit hash to vllm_ascend_test.yaml
        env:
          GITHUB_TOKEN: ${{ secrets.PAT_TOKEN }}
        run: |
          git add ./vllm_ascend_test.yaml
          git commit -s -m "[CI] Bump vllm commit hash to ${{ needs.bumper.outputs.commit_hash }}"
          git push -f origin "${{ env.BRANCH_NAME }}"

      - name: Create PR in upstream via API
        uses: actions/github-script@v8
        with:
          github-token: ${{ secrets.PAT_TOKEN }}
          script: |
            const pr = await github.rest.pulls.create({
              owner: 'vllm-project',
              repo: 'vllm-ascend',
              head: `vllm-ascend-ci:${{ env.BRANCH_NAME }}`,
              base: 'main',
              title: `[CI] Bump vllm commit hash to ${{ needs.bumper.outputs.commit_hash }}`,
              body: `This PR bumps the vllm commit hash to ${{ needs.bumper.outputs.commit_hash }} for CI purposes.`,
            });
            console.log(`Created PR #${pr.data.number}`);

[Yikun]

It seems we need to remove:

vllm-ascend/.github/workflows/format_pr_body.yaml

Lines 42 to 46 in af2a886

	- name: Get vLLM version
	working-directory: ./vllm-empty
	run: |
	VLLM_COMMIT=$(git rev-parse HEAD)
	echo "VLLM_COMMIT=https://github.com/vllm-project/vllm/commit/$VLLM_COMMIT" >> $GITHUB_ENV

and pin env.VLLM_COMMIT to the static hash

[MengqingCao]

It seems the failed cases in CI is a known issue, let's skip it

[github-actions]

This pull request has conflicts, please resolve those before we can evaluate the pull request.

[Yikun]

This patch only fix upstream interface, let's merge this to recover CI