FastDeploy/.github/workflows/ci_metax.yml

name: CI_METAX

on:
  workflow_dispatch:
  pull_request:
    types: [opened, synchronize]
    branches: [develop, release/**]

concurrency:
  group: ${{ github.event.pull_request.number }}-metax-ci
  cancel-in-progress: true

permissions: read-all

defaults:
  run:
    shell: bash

jobs:
  CI_METAX:
    runs-on: pde-ai2-squad3-fastdeploy-runner-set
    env:
      PR_ID: ${{ github.event.pull_request.number }}
      COMMIT_ID: ${{ github.event.pull_request.head.sha }}
      BRANCH: develop
    steps:
      - name: Checkout repository
        id: run-metax
        if: steps.check-bypass.outputs.can-skip != 'true'
        run: |
          export DATE_NUMBER=$(date +%Y%m%d)
          export PREV_DATE_NUMBER=$(date -d "yesterday" +%Y%m%d)

          git config --global user.name "GitHub Actions"
          git config --global user.email "actions@github.com"

          git clone https://${{ github.actor }}:${{ secrets.GITHUB_TOKEN }}@github.com/${{ github.repository }}.git .
          # git reset --hard 21f138f68be16c815f60496c1bc5ea69b511f8cc
          # git revert --no-edit 2e1680838f5e99e4ea5c5bc4251365d9add0f62f

          MODIFIED_FILES=""
          if [ "${{ github.event_name }}" == "pull_request" ]; then
            git fetch origin pull/${{ github.event.pull_request.number }}/head:pull/${{ github.event.pull_request.number }}/head
            # git rebase pull/${{ github.event.pull_request.number }}/head
            git cherry-pick FETCH_HEAD

            echo -e "\n=========== Git log info ==========="
            git --no-pager log --pretty=oneline -5
            MODIFIED_FILES=$(git --no-pager diff --name-only HEAD^ HEAD | grep -v '^$' || true)
          fi

          if [ -z "$MODIFIED_FILES" ]; then
            echo "No file change, skip metax ci."
            exit 0
          fi

          echo -e "\n=========== PR change file list ==========="
          echo "$MODIFIED_FILES"

          echo -e "\n=========== Start Metax CI Trigger Check ==========="
          echo -e "\nTarget comparison branch: remotes/origin/${BRANCH}"
          FOLLOW_PATH=(
            "custom_ops/cpu_ops/"
            "custom_ops/gpu_ops/"
            "custom_ops/metax_ops/"
            "custom_ops/*.py"
            "fastdeploy/"
            ".github/"
            "setup.py"
            "build.sh"
            "tests/ci_use/Metax_UT/"
          )
          echo -e "\nFollow path needs to be checked: ${FOLLOW_PATH[*]}"

          trigger_ci=false
          while IFS= read -r file; do
            for target_path in "${FOLLOW_PATH[@]}"; do
              if [[ $file == $target_path || $file == $target_path* ]]; then
                trigger_ci=true
                break
              fi
            done
            if [ "$trigger_ci" = true ]; then
              break
            fi
          done <<< "$MODIFIED_FILES"

          if [ "$trigger_ci" = true ]; then
            echo -e "\nExist file change found in follow path, continue metax ci."
          else
            echo -e "\nNo file change found in follow path, skip metax ci."
            exit 0
          fi

          METAX_PADDLE_CUSTOM_DEVICE_WHL_SOURCE=oss://opensource-ci/paddle
          METAX_PADDLE_CUSTOM_DEVICE_WHL_NAME=paddle_metax_gpu-3.3.0.dev${DATE_NUMBER}+maca0.0.0-cp310-cp310-linux_x86_64.whl
          # METAX_PADDLE_CUSTOM_DEVICE_WHL_NAME=paddle_metax_gpu-3.3.0.dev20251210+maca0.0.0-cp310-cp310-linux_x86_64.whl
          PADDLE_PADDLE_WHL_SOURCE=https://paddle-whl.bj.bcebos.com/nightly/cpu/paddlepaddle
          PADDLE_PADDLE_WHL_NAME=paddlepaddle-3.3.0.dev${PREV_DATE_NUMBER}-cp310-cp310-linux_x86_64.whl
          # PADDLE_PADDLE_WHL_NAME=paddlepaddle-3.3.0.dev20251209-cp310-cp310-linux_x86_64.whl

          echo -e "\n=========== Pull [ ${METAX_PADDLE_CUSTOM_DEVICE_WHL_NAME} ] from [ ${METAX_PADDLE_CUSTOM_DEVICE_WHL_SOURCE} ] ==========="
          ossutil cp ${METAX_PADDLE_CUSTOM_DEVICE_WHL_SOURCE}/${METAX_PADDLE_CUSTOM_DEVICE_WHL_NAME} .

          echo -e "\n=========== Pip install [ ${PADDLE_PADDLE_WHL_NAME} ] from [ ${PADDLE_PADDLE_WHL_SOURCE} ] ==========="
          python -m pip install ${PADDLE_PADDLE_WHL_SOURCE}/${PADDLE_PADDLE_WHL_NAME}

          echo -e "\n=========== Pip install [ ${METAX_PADDLE_CUSTOM_DEVICE_WHL_NAME} ] ==========="
          python -m pip install ${METAX_PADDLE_CUSTOM_DEVICE_WHL_NAME}

          echo -e "\n=========== Pip install [ use_triton_in_paddle ] ==========="
          python -m pip install /data/maca3.0-release-2.3/use_triton_in_paddle-0.0.0-py3-none-any.whl

      - name: Compile
        run: |
          export MACA_PATH=/opt/maca

          if [ ! -d ${HOME}/cu-bridge ]; then
                      `${MACA_PATH}/tools/cu-bridge/tools/pre_make`
          fi

          export CUDA_PATH=${HOME}/cu-bridge/CUDA_DIR
          export LD_LIBRARY_PATH=${CUDA_PATH}/lib64:${MACA_PATH}/lib:${MACA_PATH}/mxgpu_llvm/lib:$LD_LIBRARY_PATH

          PACKAGES_LINK=~/.local/lib/python3.10/site-packages
          REPLACE_FILES_PATH=/data/maca3.0-release-2.3
          cp ${REPLACE_FILES_PATH}/all_reduce.py ${PACKAGES_LINK}/paddle/distributed/communication/
          cp ${REPLACE_FILES_PATH}/all_gather.py ${PACKAGES_LINK}/paddle/distributed/communication/
          cp ${REPLACE_FILES_PATH}/broadcast.py ${PACKAGES_LINK}/paddle/distributed/communication/

          cp ${REPLACE_FILES_PATH}/cublasLt.h ${PACKAGES_LINK}/paddle/include/paddle/phi/backends/dynload/

          cp ${REPLACE_FILES_PATH}/all_things.py ${PACKAGES_LINK}/use_triton_in_paddle/cuda/

          sudo chmod 777 -R ${REPLACE_FILES_PATH}/mctlass
          sudo cp ${REPLACE_FILES_PATH}/mctlass/mctlassEx.h /opt/maca/include/mctlassEx/
          sudo cp ${REPLACE_FILES_PATH}/mctlass/libmctlassEx.so /opt/maca/lib/
          sudo cp ${REPLACE_FILES_PATH}/mctlass/mctlassEx_xcore1000.mcfb /opt/maca/lib/

          bash build.sh

          echo -e "\n=========== Pip install [ triton-3.0.0+metax3.0.0.3 ] ==========="
          python -m pip install /data/maca3.0-release-2.3/triton-3.0.0+metax3.0.0.3-cp310-cp310-linux_x86_64.whl

      - name: Run test
        run: |
          exit_code=0
          ignore_error() {
            local cmd="$*"
            echo "Execute command - [ $cmd ]"
            eval "$cmd" || {
              exit_code=$?
              echo -e "\n=========== ⚠️ Instruction execution failed (exit code $exit_code), ignore and continue. ==========="
            }
          }

          ignore_error "timeout -s 9 600s python tests/ci_use/Metax_UT/run_ernie_vl_28B.py"

          echo -e "\n=========== Fastdeploy workerlog.0 ==========="
          cat log/workerlog.0

          exit ${exit_code}