[BugFix] Fix chunked prefill (#3759) #62
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| name: Publish Job | ||
| on: | ||
| workflow_dispatch: | ||
| schedule: | ||
| - cron: '0 18 * * *' # 2:00 AM China Standard Time (UTC+8) | ||
| push: | ||
| # branches: | ||
| # - develop | ||
| tags: | ||
| - '*' | ||
| permissions: read-all | ||
| concurrency: | ||
| group: ${{ github.ref }}-${{ github.sha }} | ||
| cancel-in-progress: true | ||
| jobs: | ||
| publish_pre_check: | ||
| runs-on: ubuntu-latest | ||
| if: | | ||
| github.event.repository.fork == false && | ||
| ( | ||
| (github.event_name == 'schedule' && github.ref_name == 'develop') || | ||
| (github.event_name == 'push' && github.ref_type == 'tag') || | ||
| ((github.event_name == 'workflow_dispatch') && | ||
| (github.ref_name == 'develop' || github.ref_type == 'tag')) | ||
| ) | ||
| env: | ||
| TAG_VERSION_MAPPINGS: ${{ vars.TAG_VERSION_MAPPINGS }} | ||
| FD_VERSION_DEV: ${{ vars.FD_VERSION_DEV }} | ||
| COMPILE_USE_PADDLE_WHL_URL_MAPPINGS: ${{ vars.COMPILE_USE_PADDLE_WHL_URL_MAPPINGS }} | ||
| outputs: | ||
| compile_use_paddle_version: ${{ steps.set_output.outputs.compile_use_paddle_version }} | ||
| compile_continue: ${{ steps.set_output.outputs.compile_continue }} | ||
| fd_version: ${{ steps.set_output.outputs.fd_version }} | ||
| with_nightly_build: ${{ steps.set_output.outputs.with_nightly_build }} | ||
| compile_use_paddle_whl_url: ${{ steps.set_output.outputs.compile_use_paddle_whl_url }} | ||
| steps: | ||
| - name: Get tag version | ||
| if: github.ref_type == 'tag' | ||
| run: | | ||
| TAG_NAME="${GITHUB_REF##*/}" # 提取 tag 名称，比如 v2.1.0 | ||
| TAG_VERSION="${TAG_NAME#v}" # 去掉前缀 v | ||
| echo "FD_VERSION=$TAG_VERSION" >> $GITHUB_ENV | ||
| - name: Check FD version to Paddle version mapping | ||
| if: github.ref_type == 'tag' | ||
| env: | ||
| TARGET_FD: ${{ env.FD_VERSION }} | ||
| run: | | ||
| FOUND_PADDLE="" | ||
| # 遍历映射 | ||
| for pair in $(echo $TAG_VERSION_MAPPINGS | tr ';' ' '); do | ||
| fd=$(echo "$pair" | cut -d',' -f1) | ||
| paddle=$(echo "$pair" | cut -d',' -f2) | ||
| if [[ "$fd" == "$TARGET_FD" ]]; then | ||
| FOUND_PADDLE="$paddle" | ||
| break | ||
| fi | ||
| done | ||
| if [[ -z "$FOUND_PADDLE" ]]; then | ||
| echo "No Paddle version found for FD $TARGET_FD" | ||
| else | ||
| echo "FD $TARGET_FD maps to Paddle $FOUND_PADDLE" | ||
| echo "PADDLE_VERSION=$FOUND_PADDLE" >> $GITHUB_ENV | ||
| fi | ||
| - name: Set Version | ||
| id: set_output | ||
| env: | ||
| PADDLE_VERSION: ${{ env.PADDLE_VERSION }} | ||
| FD_VERSION: ${{ env.FD_VERSION }} | ||
| run: | | ||
| if [[ "${{ github.ref_type }}" == "tag" ]]; then | ||
| if [[ -z "$PADDLE_VERSION" ]]; then | ||
| compile_continue=false | ||
| else | ||
| compile_use_paddle_version=$PADDLE_VERSION | ||
| compile_continue=true | ||
| fi | ||
| fd_version=$FD_VERSION | ||
| fi | ||
| if [[ "${{ github.ref_name }}" == "develop" ]];then | ||
| compile_continue=true | ||
| compile_use_paddle_version="" | ||
| fd_version=${FD_VERSION_DEV} | ||
| with_nightly_build=ON | ||
| fi | ||
| # Todo | ||
| # 通过变量COMPILE_USE_PADDLE_WHL_URL_MAPPINGS中的映射关系,决定是否是安装指定版本的Paddle还是直接安装URL | ||
| for pair in $(echo $COMPILE_USE_PADDLE_WHL_URL_MAPPINGS | tr ';' ' '); do | ||
| branch=$(echo "$pair" | cut -d',' -f1) | ||
| paddle_whl_url=$(echo "$pair" | cut -d',' -f2) | ||
| if [[ "$branch" == "${{ github.ref_name }}" ]]; then | ||
| FOUND_PADDLE_URL="$paddle_whl_url" | ||
| echo "compile_use_paddle_whl_url=${FOUND_PADDLE_URL}" >> $GITHUB_OUTPUT | ||
| compile_continue=true | ||
| break | ||
| fi | ||
| done | ||
| echo "compile_continue=${compile_continue}" >> $GITHUB_OUTPUT | ||
| echo "compile_use_paddle_version=${compile_use_paddle_version}" >> $GITHUB_OUTPUT | ||
| echo "fd_version=${fd_version}" >> $GITHUB_OUTPUT | ||
| echo "with_nightly_build=${with_nightly_build:-OFF}" >> $GITHUB_OUTPUT | ||
| print_publish_pre_check_outputs: | ||
| runs-on: ubuntu-latest | ||
| needs: publish_pre_check | ||
| steps: | ||
| - name: Print outputs as JSON | ||
| run: | | ||
| echo '${{ toJSON(needs.publish_pre_check.outputs) }}' | ||
| clone: | ||
| environment: CodeSync | ||
| name: FD-Clone-Linux | ||
| runs-on: ubuntu-latest | ||
| needs: publish_pre_check | ||
| if: ${{ needs.publish_pre_check.outputs.compile_continue == 'true' }} | ||
| outputs: | ||
| repo_archive_url: ${{ steps.set_output.outputs.repo_archive_url }} | ||
| steps: | ||
| - name: Clone FastDeploy | ||
| uses: actions/checkout@v4 | ||
| with: | ||
| ref: ${{ github.ref_name }} | ||
| submodules: 'recursive' | ||
| fetch-depth: 1000 | ||
| - name: Python Setup | ||
| uses: actions/setup-python@v5 | ||
| with: | ||
| python-version: '3.10' | ||
| - name: Code Info Show and Upload | ||
| id: set_output | ||
| env: | ||
| AK: ${{ secrets.BOS_AK }} | ||
| SK: ${{ secrets.BOS_SK }} | ||
| run: | | ||
| git config --unset http.https://github.com/.extraheader | ||
| git submodule foreach --recursive sh -c "git config --local --unset-all 'http.https://github.com/.extraheader'" | ||
| git submodule foreach --recursive sh -c "git config remote.origin.fetch '+refs/heads/*:refs/remotes/origin/*'" | ||
| echo "Current HEAD Log:" | ||
| git log --oneline -n 5 | ||
| ls | ||
| cd .. | ||
| tar -zcf FastDeploy.tar.gz FastDeploy | ||
| if [[ "${{ github.ref_type }}" == "tag" ]]; then | ||
| commit_id=${{ github.sha }} | ||
| tag_name=${{ github.ref_name }} | ||
| target_path=paddle-qa/TAG/FastDeploy/${tag_name}/${commit_id} | ||
| else | ||
| commit_id=${{ github.sha }} | ||
| branch_name=${{ github.ref_name }} | ||
| target_path=paddle-qa/BRANCH/FastDeploy/${branch_name}/${commit_id} | ||
| fi | ||
| wget -q --no-proxy --no-check-certificate https://paddle-qa.bj.bcebos.com/CodeSync/develop/PaddlePaddle/PaddleTest/tools/bos_tools.py | ||
| push_file=$(realpath bos_tools.py) | ||
| python -m pip install bce-python-sdk==0.9.29 | ||
| ls | ||
| python ${push_file} FastDeploy.tar.gz ${target_path} | ||
| target_path_stripped="${target_path#paddle-qa/}" | ||
| REPO_ARCHIVE_URL=https://paddle-qa.bj.bcebos.com/${target_path_stripped}/FastDeploy.tar.gz | ||
| echo "repo_archive_url=${REPO_ARCHIVE_URL}" >> $GITHUB_OUTPUT | ||
| resultshow: | ||
| name: Show Code Archive Output | ||
| needs: clone | ||
| runs-on: ubuntu-latest | ||
| steps: | ||
| - name: Print wheel path | ||
| run: | | ||
| echo "The code archive is located at: ${{ needs.clone.outputs.repo_archive_url }}" | ||
| build_sm8090: | ||
| name: BUILD_SM8090 | ||
| needs: [clone, publish_pre_check] | ||
| uses: ./.github/workflows/_build_linux.yml | ||
| with: | ||
| DOCKER_IMAGE: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleqa:fastdeploy-ciuse-cuda126-dailyupdate | ||
| FASTDEPLOY_ARCHIVE_URL: ${{ needs.clone.outputs.repo_archive_url }} | ||
| COMPILE_ARCH: "80,90" | ||
| WITH_NIGHTLY_BUILD: ${{ needs.publish_pre_check.outputs.with_nightly_build }} | ||
| FD_VERSION: ${{ needs.publish_pre_check.outputs.fd_version }} | ||
| PADDLEVERSION: ${{ needs.publish_pre_check.outputs.compile_use_paddle_version }} | ||
| PADDLE_WHL_URL: ${{ needs.publish_pre_check.outputs.compile_use_paddle_whl_url }} | ||
| build_sm8689: | ||
| name: BUILD_SM8689 | ||
| needs: [clone, publish_pre_check] | ||
| uses: ./.github/workflows/_build_linux.yml | ||
| with: | ||
| DOCKER_IMAGE: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleqa:fastdeploy-ciuse-cuda126-dailyupdate | ||
| FASTDEPLOY_ARCHIVE_URL: ${{ needs.clone.outputs.repo_archive_url }} | ||
| COMPILE_ARCH: "86,89" | ||
| WITH_NIGHTLY_BUILD: ${{ needs.publish_pre_check.outputs.with_nightly_build }} | ||
| FD_VERSION: ${{ needs.publish_pre_check.outputs.fd_version }} | ||
| PADDLEVERSION: ${{ needs.publish_pre_check.outputs.compile_use_paddle_version }} | ||
| PADDLE_WHL_URL: ${{ needs.publish_pre_check.outputs.compile_use_paddle_whl_url }} | ||
| paddle_pypi_upload_sm8090: | ||
| environment: PaddleSourceUpload | ||
| name: PADDLE_PYPI_UPLOAD_8090 | ||
| needs: build_sm8090 | ||
| runs-on: ubuntu-latest | ||
| env: | ||
| AK: ${{ secrets.BOS_AK }} | ||
| SK: ${{ secrets.BOS_SK }} | ||
| FASTDEPLOY_WHEEL_URL: ${{ needs.build_sm8090.outputs.wheel_path }} | ||
| COMPILE_ARCH: "80,90" | ||
| steps: | ||
| - uses: actions/setup-python@v5 | ||
| with: | ||
| python-version: '3.10' | ||
| - name: Wheel Info Show and Upload | ||
| if: github.ref_name == 'develop' || github.ref_type == 'tag' | ||
| run: | | ||
| echo "The wheel is located at: ${FASTDEPLOY_WHEEL_URL}" | ||
| wget -q --no-check-certificate ${FASTDEPLOY_WHEEL_URL} | ||
| filename=$(basename ${FASTDEPLOY_WHEEL_URL}) | ||
| if [[ "${{ github.ref_name }}" == "develop" ]];then | ||
| target_path=paddle-whl/nightly/fastdeploy-gpu-${COMPILE_ARCH//,/_}/fastdeploy-gpu | ||
| elif [[ "${{ github.ref_type }}" == "tag" ]]; then | ||
| target_path=paddle-whl/stable/fastdeploy-gpu-${COMPILE_ARCH//,/_}/fastdeploy-gpu | ||
| else | ||
| echo "Not develop or tag, do nothing" | ||
| fi | ||
| wget -q --no-proxy --no-check-certificate https://paddle-qa.bj.bcebos.com/CodeSync/develop/PaddlePaddle/PaddleTest/tools/bos_tools.py | ||
| push_file=$(realpath bos_tools.py) | ||
| python -m pip install bce-python-sdk==0.9.29 | ||
| ls | ||
| python ${push_file} ${filename} ${target_path} | ||
| paddle_pypi_upload_sm8689: | ||
| environment: PaddleSourceUpload | ||
| name: PADDLE_PYPI_UPLOAD_8689 | ||
| needs: build_sm8689 | ||
| runs-on: ubuntu-latest | ||
| env: | ||
| AK: ${{ secrets.BOS_AK }} | ||
| SK: ${{ secrets.BOS_SK }} | ||
| FASTDEPLOY_WHEEL_URL: ${{ needs.build_sm8689.outputs.wheel_path }} | ||
| COMPILE_ARCH: "86,89" | ||
| steps: | ||
| - uses: actions/setup-python@v5 | ||
| with: | ||
| python-version: '3.10' | ||
| - name: Wheel Info Show and Upload | ||
| if: github.ref_name == 'develop' || github.ref_type == 'tag' | ||
| run: | | ||
| echo "The wheel is located at: ${FASTDEPLOY_WHEEL_URL}" | ||
| wget -q --no-check-certificate ${FASTDEPLOY_WHEEL_URL} | ||
| filename=$(basename ${FASTDEPLOY_WHEEL_URL}) | ||
| if [[ "${{ github.ref_name }}" == "develop" ]];then | ||
| target_path=paddle-whl/nightly/fastdeploy-gpu-${COMPILE_ARCH//,/_}/fastdeploy-gpu | ||
| elif [[ "${{ github.ref_type }}" == "tag" ]]; then | ||
| target_path=paddle-whl/stable/fastdeploy-gpu-${COMPILE_ARCH//,/_}/fastdeploy-gpu | ||
| else | ||
| echo "Not develop or tag, do nothing" | ||
| fi | ||
| wget -q --no-proxy --no-check-certificate https://paddle-qa.bj.bcebos.com/CodeSync/develop/PaddlePaddle/PaddleTest/tools/bos_tools.py | ||
| push_file=$(realpath bos_tools.py) | ||
| python -m pip install bce-python-sdk==0.9.29 | ||
| ls | ||
| python ${push_file} ${filename} ${target_path} | ||
| unittest_coverage: | ||
| name: Run FastDeploy Unit Tests and Coverage | ||
| needs: [clone,build_sm8090] | ||
| uses: ./.github/workflows/_unit_test_coverage.yml | ||
| with: | ||
| DOCKER_IMAGE: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleqa:fastdeploy-ciuse-cuda126-dailyupdate | ||
| FASTDEPLOY_ARCHIVE_URL: ${{ needs.clone.outputs.repo_archive_url }} | ||
| FASTDEPLOY_WHEEL_URL: ${{ needs.build_sm8090.outputs.wheel_path }} | ||
| MODEL_CACHE_DIR: "/ssd2/actions-runner/ModelData" | ||
| secrets: | ||
| github-token: ${{ secrets.GITHUB_TOKEN }} | ||
| logprob_test: | ||
| name: Run FastDeploy LogProb Tests | ||
| needs: [build_sm8090] | ||
| uses: ./.github/workflows/_logprob_test_linux.yml | ||
| with: | ||
| DOCKER_IMAGE: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleqa:fastdeploy-ciuse-cuda126-dailyupdate | ||
| PADDLETEST_ARCHIVE_URL: "https://xly-devops.bj.bcebos.com/PaddleTest/PaddleTest.tar.gz" | ||
| FASTDEPLOY_WHEEL_URL: ${{ needs.build_sm8090.outputs.wheel_path }} | ||
| MODEL_CACHE_DIR: "/ssd2/actions-runner/ModelData" | ||
| pre_ce_test: | ||
| name: Extracted partial CE model tasks to run in CI. | ||
| needs: [clone,build_sm8090] | ||
| uses: ./.github/workflows/_pre_ce_test.yml | ||
| with: | ||
| DOCKER_IMAGE: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleqa:fastdeploy-ciuse-cuda126-dailyupdate | ||
| FASTDEPLOY_ARCHIVE_URL: ${{ needs.clone.outputs.repo_archive_url }} | ||
| FASTDEPLOY_WHEEL_URL: ${{ needs.build_sm8090.outputs.wheel_path }} | ||
| MODEL_CACHE_DIR: "/ssd2/actions-runner/ModelData" | ||
| base_test: | ||
| name: Run Base Tests | ||
| needs: [clone,build_sm8090] | ||
| uses: ./.github/workflows/_base_test.yml | ||
| with: | ||
| DOCKER_IMAGE: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleqa:fastdeploy-ciuse-cuda126-dailyupdate | ||
| FASTDEPLOY_ARCHIVE_URL: ${{ needs.clone.outputs.repo_archive_url }} | ||
| FASTDEPLOY_WHEEL_URL: ${{ needs.build_sm8090.outputs.wheel_path }} | ||
| MODEL_CACHE_DIR: "/ssd2/actions-runner/ModelData" | ||
| accuracy_test: | ||
| name: Run Accuracy Tests | ||
| needs: [clone,build_sm8090] | ||
| uses: ./.github/workflows/_accuracy_test.yml | ||
| with: | ||
| DOCKER_IMAGE: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleqa:fastdeploy-ciuse-cuda126-dailyupdate | ||
| FASTDEPLOY_ARCHIVE_URL: ${{ needs.clone.outputs.repo_archive_url }} | ||
| FASTDEPLOY_WHEEL_URL: ${{ needs.build_sm8090.outputs.wheel_path }} | ||
| MODEL_CACHE_DIR: "/ssd2/actions-runner/ModelData" | ||
| stable_test: | ||
| name: Run Stable Tests | ||
| needs: [clone,build] | ||
| uses: ./.github/workflows/_stable_test.yml | ||
| with: | ||
| DOCKER_IMAGE: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleqa:fastdeploy-ciuse-cuda126-dailyupdate | ||
| FASTDEPLOY_ARCHIVE_URL: ${{ needs.clone.outputs.repo_archive_url }} | ||
| FASTDEPLOY_WHEEL_URL: ${{ needs.build.outputs.wheel_path }} | ||
| MODEL_CACHE_DIR: "/ssd2/actions-runner/ModelData" | ||
