Add harness nightly #101
Workflow file for this run
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 Harness Evalution | |
| # 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 | |
| # Controls when the action will run. | |
| on: | |
| schedule: | |
| - cron: "00 13 * * 5" # GMT time, 13:00 GMT == 21:00 China | |
| pull_request: | |
| branches: [main] | |
| paths: | |
| - ".github/workflows/llm-harness-evaluation.yml" | |
| # Allows you to run this workflow manually from the Actions tab | |
| workflow_dispatch: | |
| # A workflow run is made up of one or more jobs that can run sequentially or in parallel | |
| jobs: | |
| llm-cpp-build: | |
| uses: ./.github/workflows/llm-binary-build.yml | |
| llm-harness-evalution: | |
| timeout-minutes: 1000 | |
| needs: llm-cpp-build | |
| strategy: | |
| fail-fast: false | |
| matrix: | |
| # include: | |
| # python-version: "3.9" | |
| # model_name: "stablelm-3b-4e1t" | |
| # task: "arc" | |
| # precision: "sym_int4" | |
| python-version: ["3.9"] | |
| model_name: [stablelm-3b-4e1t] | |
| task: [truthfulqa] | |
| precision: [mixed_fp4] | |
| device: [xpu] | |
| runs-on: [self-hosted, llm, temp-arc01] | |
| env: | |
| ANALYTICS_ZOO_ROOT: ${{ github.workspace }} | |
| ORIGIN_DIR: /mnt/disk1/models | |
| HARNESS_HF_HOME: /mnt/disk1/harness_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 | |
| - name: Run LLM install (all) test | |
| uses: ./.github/actions/llm/setup-llm-env | |
| with: | |
| extra-dependency: "xpu" | |
| - name: Install harness | |
| working-directory: ${{ github.workspace }}/python/llm/dev/benchmark/harness/ | |
| shell: bash | |
| run: | | |
| git clone https://github.com/EleutherAI/lm-evaluation-harness.git | |
| cd lm-evaluation-harness | |
| git checkout e81d3cc | |
| pip install -e . | |
| - name: Download models and datasets | |
| shell: bash | |
| run: | | |
| echo "MODEL_PATH=${ORIGIN_DIR}/${{ matrix.model_name }}/" >> "$GITHUB_ENV" | |
| MODEL_PATH=${ORIGIN_DIR}/${{ matrix.model_name }}/ | |
| if [ ! -d $HARNESS_HF_HOME ]; then | |
| mkdir -p $HARNESS_HF_HOME | |
| wget -r -nH -l inf --no-verbose --cut-dirs=2 ${LLM_FTP_URL}/llm/LeaderBoard_Datasets/ -P $HARNESS_HF_HOME/ | |
| fi | |
| if [ ! -d $MODEL_PATH ]; then | |
| wget -r -nH --no-verbose --cut-dirs=1 ${LLM_FTP_URL}/llm/${{ matrix.model_name }} -P ${ORIGIN_DIR} | |
| fi | |
| - name: Upgrade packages | |
| shell: bash | |
| run: | | |
| pip install --upgrade transformers==4.34.0 | |
| - name: Run harness | |
| shell: bash | |
| working-directory: ${{ github.workspace }}/python/llm/dev/benchmark/harness | |
| env: | |
| USE_XETLA: OFF | |
| # SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS: 1 | |
| run: | | |
| export HF_HOME=${HARNESS_HF_HOME} | |
| export HF_DATASETS=$HARNESS_HF_HOME/datasets | |
| export HF_DATASETS_CACHE=$HARNESS_HF_HOME/datasets | |
| source /opt/intel/oneapi/setvars.sh | |
| python run_llb.py --model bigdl-llm --pretrained ${MODEL_PATH} --precision ${{ matrix.precision }} --device ${{ matrix.device }} --tasks ${{ matrix.task }} --batch_size 1 --no_cache --output_path results | |
| - name: Compare with golden accuracy | |
| shell: bash | |
| working-directory: ${{ github.workspace }}/python/llm/dev/benchmark/harness | |
| run: | | |
| echo '{' >> golden_results.json | |
| echo '"stablelm-3b-4e1t": {"xpu": {"mixed_fp4": {"truthfulqa_mc": {"mc1": 0.4,"mc1_stderr": 0.16329931618554522,"mc2": 0.46830478972624556,"mc2_stderr": 0.119}, "arc_challenge": {"acc": 0.5, "acc_stderr": 0.5, "acc_norm": 0.5, "acc_norm_stderr": 0.5}}}},' >> golden_results.json | |
| echo '"stablelm-3b-4e1t": {"xpu": {"fp8": {"truthfulqa_mc": {"mc1": 0.4,"mc1_stderr": 0.16329931618554522,"mc2": 0.46830478972624556,"mc2_stderr": 0.119}, "arc_challenge": {"acc": 0.5, "acc_stderr": 0.5, "acc_norm": 0.5, "acc_norm_stderr": 0.5}}}},' >> golden_results.json | |
| echo '"Mistral-7B-v0.1": {"xpu": {"mixed_fp4": {"truthfulqa_mc": {"mc1": 0.27539779681762544,"mc1_stderr": 0.01563813566777552,"mc2": 0.41062244273774384,"mc2_stderr": 0.014067078150027909}, "arc_challenge": {"acc": 0.5674061433447098,"acc_stderr": 0.014478005694182528,"acc_norm": 0.5989761092150171,"acc_norm_stderr": 0.014322255790719867}}}},' >> golden_results.json | |
| echo '"Mistral-7B-v0.1": {"xpu": {"fp8": {"truthfulqa_mc": {"mc1": 0.2778457772337821,"mc1_stderr": 0.015680929364024643,"mc2": 0.4212635093545362,"mc2_stderr": 0.01414660694632397}, "arc_challenge": {"acc": 0.5639931740614335,"acc_stderr": 0.014491225699230916, "acc_norm": 0.5998293515358362,"acc_norm_stderr": 0.014317197787809174}}}}' >> golden_results.json | |
| echo '}' >> golden_results.json | |
| python accuracy_regression.py results/${{ matrix.model_name }}/${{ matrix.device }}/${{ matrix.precision }}/${{ matrix.task }}/result.json golden_results.json | |