LLM C-Eval #176
This file contains 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: LLM C-Eval | |
| # Cancel previous runs in the PR when you push new commits | |
| concurrency: | |
| group: ${{ github.workflow }}-llm-nightly-test-${{ github.event.pull_request.number || github.run_id }} | |
| cancel-in-progress: true | |
| permissions: | |
| contents: read | |
| # Controls when the action will run. | |
| on: | |
| schedule: | |
| - cron: "00 15 * * *" # GMT time, 15:00 GMT == 23:00 Beijing Time | |
| pull_request: | |
| branches: [main] | |
| paths: | |
| - ".github/workflows/llm-c-evaluation.yml" | |
| # Allows you to run this workflow manually from the Actions tab | |
| workflow_dispatch: | |
| inputs: | |
| model_name: | |
| description: 'Model names, separated by comma and must be quoted.' | |
| required: true | |
| type: string | |
| precision: | |
| description: 'Precisions, separated by comma and must be quoted.' | |
| required: true | |
| type: string | |
| runs-on: | |
| description: 'Labels to filter the runners, separated by comma and must be quoted.' | |
| default: "accuracy" | |
| required: false | |
| type: string | |
| # A workflow run is made up of one or more jobs that can run sequentially | |
| jobs: | |
| llm-cpp-build: | |
| uses: ./.github/workflows/llm-binary-build.yml | |
| with: | |
| platform: 'Linux' | |
| # Set the testing matrix based on the event (schedule, PR, or manual dispatch) | |
| set-matrix: | |
| runs-on: ubuntu-latest | |
| outputs: | |
| model_name: ${{ steps.set-matrix.outputs.model_name }} | |
| precision: ${{ steps.set-matrix.outputs.precision }} | |
| runner: ${{ steps.set-matrix.outputs.runner }} | |
| date: ${{ steps.set-matrix.outputs.date }} | |
| steps: | |
| - name: set-nightly-env | |
| if: ${{github.event_name == 'schedule'}} | |
| env: | |
| NIGHTLY_MATRIX_MODEL_NAME: '["chatglm2-6b","chinese-llama2-7b", "Qwen-7B-Chat"]' | |
| NIGHTLY_MATRIX_PRECISION: '["sym_int4", "fp8_e5m2"]' | |
| NIGHTLY_LABELS: '["self-hosted", "llm", "accuracy-nightly"]' | |
| run: | | |
| echo "model_name=$NIGHTLY_MATRIX_MODEL_NAME" >> $GITHUB_ENV | |
| echo "precision=$NIGHTLY_MATRIX_PRECISION" >> $GITHUB_ENV | |
| echo "runner=$NIGHTLY_LABELS" >> $GITHUB_ENV | |
| - name: set-pr-env | |
| if: ${{github.event_name == 'pull_request'}} | |
| env: | |
| PR_MATRIX_MODEL_NAME: '["Qwen-7B-Chat"]' | |
| PR_MATRIX_PRECISION: '["sym_int4", "fp8_e5m2"]' | |
| PR_LABELS: '["self-hosted", "llm", "temp-arc01"]' | |
| run: | | |
| echo "model_name=$PR_MATRIX_MODEL_NAME" >> $GITHUB_ENV | |
| echo "precision=$PR_MATRIX_PRECISION" >> $GITHUB_ENV | |
| echo "runner=$PR_LABELS" >> $GITHUB_ENV | |
| - name: set-manual-env | |
| if: ${{github.event_name == 'workflow_dispatch'}} | |
| env: | |
| MANUAL_MATRIX_MODEL_NAME: ${{format('[ {0} ]', inputs.model_name)}} | |
| MANUAL_MATRIX_PRECISION: ${{format('[ {0} ]', inputs.precision)}} | |
| MANUAL_LABELS: ${{format('["self-hosted", "llm", {0}]', inputs.runs-on)}} | |
| run: | | |
| echo "model_name=$MANUAL_MATRIX_MODEL_NAME" >> $GITHUB_ENV | |
| echo "precision=$MANUAL_MATRIX_PRECISION" >> $GITHUB_ENV | |
| echo "runner=$MANUAL_LABELS" >> $GITHUB_ENV | |
| - name: set-matrix | |
| id: set-matrix | |
| run: | | |
| echo "model_name=$model_name" >> $GITHUB_OUTPUT | |
| echo "precision=$precision" >> $GITHUB_OUTPUT | |
| echo "runner=$runner" >> $GITHUB_OUTPUT | |
| echo "date=$(date +%Y-%m-%d)" >> $GITHUB_OUTPUT | |
| llm-c-evaluation: | |
| timeout-minutes: 1200 | |
| needs: [llm-cpp-build, set-matrix] | |
| strategy: | |
| fail-fast: false | |
| matrix: | |
| python-version: ["3.11"] | |
| model_name: ${{ fromJson(needs.set-matrix.outputs.model_name) }} | |
| precision: ${{ fromJson(needs.set-matrix.outputs.precision) }} | |
| device: [xpu] | |
| runs-on: ${{ fromJson(needs.set-matrix.outputs.runner) }} | |
| env: | |
| ANALYTICS_ZOO_ROOT: ${{ github.workspace }} | |
| ORIGIN_DIR: /mnt/disk1/models | |
| CEVAL_HF_HOME: /mnt/disk1/ceval_home | |
| steps: | |
| - uses: actions/checkout@v3 | |
| - name: Set up Python ${{ matrix.python-version }} | |
| uses: actions/setup-python@v4 | |
| with: | |
| python-version: ${{ matrix.python-version }} | |
| - name: Install dependencies | |
| shell: bash | |
| run: | | |
| python -m pip install --upgrade pip | |
| python -m pip install --upgrade setuptools==58.0.4 | |
| python -m pip install --upgrade wheel | |
| - name: Download llm binary | |
| uses: ./.github/actions/llm/download-llm-binary | |
| with: | |
| platform: 'Linux' | |
| - name: Run LLM install (all) test | |
| uses: ./.github/actions/llm/setup-llm-env | |
| with: | |
| extra-dependency: "xpu_2.1" | |
| - name: Download models and datasets | |
| shell: bash | |
| run: | | |
| echo "MODEL_PATH=${ORIGIN_DIR}/${{ matrix.model_name }}/" >> "$GITHUB_ENV" | |
| MODEL_PATH=${ORIGIN_DIR}/${{ matrix.model_name }}/ | |
| if [ ! -d $CEVAL_HF_HOME ]; then | |
| mkdir -p $CEVAL_HF_HOME | |
| fi | |
| if [ ! -d "$CEVAL_HF_HOME/data" ]; then | |
| mkdir -p "$CEVAL_HF_HOME/data" | |
| fi | |
| if [ -d "$CEVAL_HF_HOME/data/dev" ]; then | |
| rm -rf "$CEVAL_HF_HOME/data/dev" | |
| fi | |
| if [ -d "$CEVAL_HF_HOME/data/test" ]; then | |
| rm -rf "$CEVAL_HF_HOME/data/test" | |
| fi | |
| if [ -d "$CEVAL_HF_HOME/data/val" ]; then | |
| rm -rf "$CEVAL_HF_HOME/data/val" | |
| fi | |
| wget -r -nH -nc --no-verbose --cut-dirs=1 ${LLM_FTP_URL}/llm/ceval-exam.zip -P "$CEVAL_HF_HOME/data" | |
| echo "DATA_PATH=$CEVAL_HF_HOME/data" >> "$GITHUB_ENV" | |
| DATA_PATH=$CEVAL_HF_HOME/data | |
| unzip -o "$CEVAL_HF_HOME/data/ceval-exam.zip" -d "$CEVAL_HF_HOME/data" | |
| wget -r -nH -nc --no-verbose --cut-dirs=1 ${LLM_FTP_URL}/llm/${{ matrix.model_name }} -P ${ORIGIN_DIR} | |
| - name: Install New Dependencies | |
| shell: bash | |
| run: | | |
| pip install einops | |
| pip install thefuzz | |
| pip install tiktoken | |
| pip install numpy==1.26.3 | |
| pip install pandas==1.5.3 | |
| pip install transformers_stream_generator | |
| - name: Run C-Eval | |
| shell: bash | |
| working-directory: ${{ github.workspace }}/python/llm/dev/benchmark/ceval | |
| env: | |
| USE_XETLA: OFF | |
| SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS: 1 | |
| run: | | |
| source /opt/intel/oneapi/setvars.sh | |
| python eval.py \ | |
| --model_path ${MODEL_PATH} \ | |
| --eval_type validation \ | |
| --device xpu \ | |
| --eval_data_path ${DATA_PATH} \ | |
| --qtype ${{ matrix.precision }} | |
| - uses: actions/upload-artifact@v3 | |
| with: | |
| name: ceval_results | |
| path: | |
| ${{ github.workspace }}/python/llm/dev/benchmark/ceval/results/** | |
| llm-ceval-summary: | |
| if: ${{ always() }} | |
| needs: [set-matrix, llm-c-evaluation] | |
| runs-on: ubuntu-latest | |
| steps: | |
| - uses: actions/checkout@v3 | |
| - name: Set up Python 3.11 | |
| uses: actions/setup-python@v4 | |
| with: | |
| python-version: 3.11 | |
| - name: Install dependencies | |
| shell: bash | |
| run: | | |
| pip install --upgrade pip | |
| pip install pandas==1.5.3 | |
| - name: Download ceval results | |
| uses: actions/download-artifact@v3 | |
| with: | |
| name: ceval_results | |
| path: results | |
| - name: Summarize the results | |
| shell: bash | |
| run: | | |
| ls results | |
| echo "DATE=$(date +%Y-%m-%d)" >> $GITHUB_ENV | |
| python ${{ github.workspace }}/python/llm/dev/benchmark/ceval/organize_results.py results/ results/ | |
| - name: Set artifact file path | |
| run: echo "ARTIFACT_PATH=results/results_${{ needs.set-matrix.outputs.date }}.csv" >> $GITHUB_ENV | |
| - uses: actions/upload-artifact@v3 | |
| with: | |
| name: results_${{ needs.set-matrix.outputs.date }} | |
| path: ${{ env.ARTIFACT_PATH }} | |
| llm-ceval-html: | |
| if: ${{github.event_name == 'schedule' || github.event_name == 'pull_request'}} | |
| needs: [set-matrix, llm-c-evaluation, llm-ceval-summary] | |
| runs-on: ["self-hosted", "llm", "accuracy1", "accuracy-nightly"] | |
| steps: | |
| - uses: actions/checkout@f43a0e5ff2bd294095638e18286ca9a3d1956744 # actions/checkout@v3 | |
| - name: Set up Python 3.11 | |
| uses: actions/setup-python@v4 | |
| with: | |
| python-version: 3.11 | |
| - name: Install dependencies | |
| shell: bash | |
| run: | | |
| pip install --upgrade pip | |
| pip install numpy | |
| pip install pandas==1.5.3 | |
| pip install jsonlines pytablewriter regex | |
| - name: Set output path | |
| shell: bash | |
| run: | | |
| if ${{github.event_name == 'pull_request'}}; then | |
| echo 'ACC_FOLDER=/home/arda/action-runners/ceval/pr-accuracy-data' >> $GITHUB_ENV | |
| fi | |
| if ${{github.event_name == 'schedule'}}; then | |
| echo 'ACC_FOLDER=/home/arda/action-runners/ceval/nightly-accuracy-data' >> $GITHUB_ENV | |
| fi | |
| - name: Create ceval results directory if not exists | |
| run: | | |
| if [ ! -d "${{ env.ACC_FOLDER }}" ]; then | |
| mkdir -p "${{ env.ACC_FOLDER }}" | |
| fi | |
| - name: Download ceval results | |
| uses: actions/download-artifact@v3 | |
| with: | |
| name: results_${{ needs.set-matrix.outputs.date }} | |
| path: ${{ env.ACC_FOLDER }} | |
| rename: results_${{ needs.set-matrix.outputs.date }}.csv | |
| - name: Update HTML | |
| working-directory: ${{ github.workspace }}/python/llm/test/benchmark/ceval/ | |
| shell: bash | |
| run: | | |
| python ceval_csv_to_html.py -f $ACC_FOLDER | |
| if ${{github.event_name == 'schedule'}}; then | |
| python ../harness/update_html_in_parent_folder.py -f $ACC_FOLDER | |
| fi |