xc-llm-ascend/.github/workflows/_accuracy_test.yaml

name: 'accuracy test'

on:
  workflow_call:
    inputs:
      vllm:
        required: true
        type: string
      vllm-ascend:
        required: false
        type: string
        default: main
      runner:
        required: true
        type: string
      image:
        required: true
        type: string
      model_name:
        required: true
        type: string
      upload:
        required: false
        type: boolean
        default: false

jobs:
  accuracy_tests:

    runs-on: ${{ inputs.runner }}
    name: ${{ inputs.model_name }} accuracy
    container:
      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.3.rc2-910b-ubuntu22.04-py3.11
      env:
        VLLM_USE_MODELSCOPE: True
        # 1. If version specified (work_dispatch), do specified branch accuracy test
        # 2. If no version (labeled PR), do accuracy test by default ref:
        # The branch, tag or SHA to checkout. When checking out the repository that
        # triggered a workflow, this defaults to the reference or SHA for that event.
        # Otherwise, uses the default branch.
        GHA_VLLM_ASCEND_VERSION: ${{ inputs.vllm-ascend }}

    steps:
      - name: Checkout repository
        uses: actions/checkout@v4

      - name: Set model name as output
        id: set_output
        run: |
          echo "model_name=${{ inputs.model_name }}" >> $GITHUB_OUTPUT

      - name: Config mirrors
        run: |
          sed -Ei 's@(ports|archive).ubuntu.com@cache-service.nginx-pypi-cache.svc.cluster.local:8081@g' /etc/apt/sources.list
          pip config set global.index-url http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple
          pip config set global.trusted-host cache-service.nginx-pypi-cache.svc.cluster.local
          apt-get update -y
          apt install git -y

      - name: Install system dependencies
        run: |
          apt-get -y install `cat packages.txt`
          apt-get -y install gcc g++ cmake libnuma-dev

      - name: Checkout vllm-project/vllm repo
        uses: actions/checkout@v4
        with:
          repository: vllm-project/vllm
          ref: ${{ inputs.vllm }}
          path: ./vllm-empty

      - name: Install vllm-project/vllm from source
        working-directory: ./vllm-empty
        run: |
          VLLM_TARGET_DEVICE=empty pip install -e .

      - name: Resolve vllm-ascend version
        run: |
          VERSION_INPUT="${{ inputs.vllm-ascend }}"
          
          if [[ "$VERSION_INPUT" == "latest" ]]; then
            TAGS=$(git ls-remote --tags --sort=-v:refname https://github.com/vllm-project/vllm-ascend "v*" | cut -f2 | sed 's|refs/tags/||')
            LATEST_TAG=$(echo "$TAGS" | head -n1)
            if [[ -z "$LATEST_TAG" ]]; then
              RESOLVED_VERSION="main"
            else
              RESOLVED_VERSION="$LATEST_TAG"
            fi
          else
            RESOLVED_VERSION="$VERSION_INPUT"
          fi
          echo "GHA_VLLM_ASCEND_VERSION=$RESOLVED_VERSION" >> $GITHUB_ENV

      - name: Checkout vllm-project/vllm-ascend repo
        uses: actions/checkout@v4
        with:
          repository: vllm-project/vllm-ascend
          path: ./vllm-ascend
          ref: ${{ env.GHA_VLLM_ASCEND_VERSION }}

      - name: Install vllm-project/vllm-ascend
        working-directory: ./vllm-ascend
        env:
          PIP_EXTRA_INDEX_URL: https://mirrors.huaweicloud.com/ascend/repos/pypi
        run: |
          pip install -r requirements-dev.txt
          pip install -v -e .

      - name: Get vLLM commit hash and URL
        working-directory: ./vllm-empty
        run: |
          VLLM_COMMIT=$(git rev-parse --short=7 HEAD)
          echo "VLLM_COMMIT=$VLLM_COMMIT" >> $GITHUB_ENV

      - name: Get vLLM-Ascend commit hash and URL
        working-directory: ./vllm-ascend
        run: |
          VLLM_ASCEND_COMMIT=$(git rev-parse --short=7 HEAD)
          echo "VLLM_ASCEND_COMMIT=$VLLM_ASCEND_COMMIT" >> $GITHUB_ENV

      - name: Collect version info
        run: |
          for dir in /usr/local/Ascend/ascend-toolkit/*; do
            dname=$(basename "$dir")
            if [ "$dname" != "latest" ]; then
              TOOLKIT_DIR="$dname"
              break
            fi
          done
          INFO_FILE="/usr/local/Ascend/ascend-toolkit/${TOOLKIT_DIR}/$(uname -i)-linux/ascend_toolkit_install.info"
          GHA_CANN_VERSION=$(grep "version=" "$INFO_FILE" \
                           | head -n1 \
                           | cut -d'=' -f2 \
                           | tr -d '"')
          {
            echo "GHA_CANN_VERSION=$GHA_CANN_VERSION"
            pip show torch | grep "Version:" | awk '{print "GHA_TORCH_VERSION="$2}'
            pip show torch_npu | grep "Version:" | awk '{print "GHA_TORCH_NPU_VERSION="$2}'
            pip show vllm | grep "Version:" | awk '{print "GHA_VLLM_VERSION="$2}' | sed 's/+.*//'
          } >> "$GITHUB_ENV"

      - name: Run accuracy test
        id: report
        env:
          VLLM_WORKER_MULTIPROC_METHOD: spawn
          VLLM_USE_MODELSCOPE: True
          VLLM_VERSION: ${{ env.GHA_VLLM_VERSION }}
          VLLM_COMMIT: ${{ env.VLLM_COMMIT }}
          VLLM_ASCEND_VERSION: ${{ env.GHA_VLLM_ASCEND_VERSION || github.ref }}
          VLLM_ASCEND_COMMIT: ${{ env.VLLM_ASCEND_COMMIT }}
          CANN_VERSION: ${{ env.GHA_CANN_VERSION }}
          TORCH_VERSION: ${{ env.GHA_TORCH_VERSION }}
          TORCH_NPU_VERSION: ${{ env.GHA_TORCH_NPU_VERSION }}
        run: |
          model_base_name=$(basename ${{ inputs.model_name }})
          markdown_name="${model_base_name}"
          echo "markdown_name=$markdown_name" >> $GITHUB_OUTPUT
          mkdir -p ./benchmarks/accuracy
          pytest -sv ./tests/e2e/models/test_lm_eval_correctness.py \
          --config ./tests/e2e/models/configs/${{ inputs.model_name }}.yaml

      - name: Generate step summary
        if: ${{ always() }}
        run: |
          cat ./benchmarks/accuracy/${{ steps.report.outputs.markdown_name }}.md >> $GITHUB_STEP_SUMMARY

      - name: Upload Report
        if: ${{ inputs.upload == true }}
        uses: actions/upload-artifact@v4
        with:
          name: "report-${{ env.GHA_VLLM_ASCEND_VERSION }}-${{ steps.report.outputs.markdown_name }}"
          path: ./benchmarks/accuracy/${{ steps.report.outputs.markdown_name }}.md
          if-no-files-found: warn
          retention-days: 90
          overwrite: true