TorchBench V3 bisection #78
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: TorchBench V3 bisection | |
| on: | |
| workflow_dispatch: | |
| inputs: | |
| regression_date: | |
| description: "Date of the regression" | |
| required: true | |
| default: "2023-08-29" | |
| jobs: | |
| bisection: | |
| environment: docker-s3-upload | |
| env: | |
| BASE_CONDA_ENV: "torchbench" | |
| CONDA_ENV: "bisection-ci-v3" | |
| PLATFORM_NAME: "gcp_a100" | |
| USERBENCHMARK_NAME: "torch-nightly" | |
| SETUP_SCRIPT: "/workspace/setup_instance.sh" | |
| BISECT_WORKDIR: ".userbenchmark/torch-nightly/bisection" | |
| AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_ID }} | |
| AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }} | |
| if: ${{ github.repository_owner == 'pytorch' }} | |
| runs-on: [self-hosted, a100-runner] | |
| timeout-minutes: 2880 # 48 hours | |
| steps: | |
| - name: Checkout | |
| uses: actions/checkout@v3 | |
| with: | |
| path: benchmark | |
| - name: Checkout pytorch | |
| uses: actions/checkout@v3 | |
| with: | |
| repository: pytorch/pytorch | |
| path: srcs/pytorch | |
| fetch-depth: 0 | |
| - name: Checkout torchvision | |
| uses: actions/checkout@v3 | |
| with: | |
| repository: pytorch/vision | |
| path: srcs/vision | |
| fetch-depth: 0 | |
| - name: Checkout torchdata | |
| uses: actions/checkout@v3 | |
| with: | |
| repository: pytorch/data | |
| path: srcs/data | |
| fetch-depth: 0 | |
| - name: Checkout torchaudio | |
| uses: actions/checkout@v3 | |
| with: | |
| repository: pytorch/audio | |
| path: srcs/audio | |
| fetch-depth: 0 | |
| - name: Tune Nvidia GPU | |
| run: | | |
| sudo nvidia-smi -pm 1 | |
| sudo nvidia-smi -ac 1215,1410 | |
| nvidia-smi | |
| - name: Install Deps | |
| run: | | |
| sudo apt-get -y update && sudo apt -y update | |
| - name: Setup conda env | |
| run: | | |
| CONDA_ENV=${BASE_CONDA_ENV} . "${SETUP_SCRIPT}" | |
| cd benchmark | |
| python ./utils/python_utils.py --create-conda-env "${CONDA_ENV}" | |
| - name: Setup bisection environment | |
| run: | | |
| . "${SETUP_SCRIPT}"; cd benchmark | |
| python utils/cuda_utils.py --install-torch-build-deps | |
| python utils/cuda_utils.py --install-torchbench-deps | |
| mkdir -p "${BISECT_WORKDIR}" | |
| REGRESSION_DATE="${{ github.event.inputs.regression_date }}" | |
| python regression_detector.py --name "${USERBENCHMARK_NAME}" --platform "${PLATFORM_NAME}" \ | |
| --end-date "${REGRESSION_DATE}" --download-from-s3 --output "${BISECT_WORKDIR}/regression-${REGRESSION_DATE}.yaml" | |
| - name: Bisection | |
| run: | | |
| . "${SETUP_SCRIPT}"; cd benchmark | |
| REGRESSION_DATE="${{ github.event.inputs.regression_date }}" | |
| python bisection.py --work-dir "${BISECT_WORKDIR}" --torch-repos-path "${PWD}/../srcs" \ | |
| --torchbench-repo-path "${PWD}" --config "${BISECT_WORKDIR}/regression-${REGRESSION_DATE}.yaml" \ | |
| --output "${BISECT_WORKDIR}/bisect-output-gh${GITHUB_RUN_ID}.json" | |
| cp -r "${BISECT_WORKDIR}" ../bisection-result | |
| - name: Upload artifact | |
| if: always() | |
| uses: actions/upload-artifact@v3 | |
| with: | |
| name: Bisection result | |
| path: bisection-result/ | |
| - name: Clean up Conda env | |
| if: always() | |
| run: | | |
| . "${SETUP_SCRIPT}" | |
| conda deactivate && conda deactivate | |
| conda remove -n "${CONDA_ENV}" --all |