apple-perf #84
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: apple-perf | |
| on: | |
| schedule: | |
| - cron: 0 1 * * * | |
| # Note: GitHub has an upper limit of 10 inputs | |
| workflow_dispatch: | |
| inputs: | |
| models: | |
| description: Models to be benchmarked | |
| required: false | |
| type: string | |
| default: stories110M | |
| devices: | |
| description: Target devices to run benchmark | |
| required: false | |
| type: string | |
| default: apple_iphone_15 | |
| delegates: | |
| description: Backend delegates | |
| required: false | |
| type: string | |
| default: xnnpack | |
| benchmark_configs: | |
| description: The list of configs used the benchmark | |
| required: false | |
| type: string | |
| test_spec: | |
| description: The test spec to drive the test on AWS devices | |
| required: false | |
| type: string | |
| workflow_call: | |
| inputs: | |
| models: | |
| description: Models to be benchmarked | |
| required: false | |
| type: string | |
| default: stories110M | |
| devices: | |
| description: Target devices to run benchmark | |
| required: false | |
| type: string | |
| default: apple_iphone_15 | |
| delegates: | |
| description: Backend delegates | |
| required: false | |
| type: string | |
| default: xnnpack | |
| benchmark_configs: | |
| description: The list of configs used the benchmark | |
| required: false | |
| type: string | |
| test_spec: | |
| description: The test spec to drive the test on AWS devices | |
| required: false | |
| type: string | |
| concurrency: | |
| group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref_name }}-${{ github.ref_type == 'branch' && github.sha }}-${{ github.event_name == 'workflow_dispatch' }}-${{ github.event_name == 'schedule' }} | |
| cancel-in-progress: true | |
| jobs: | |
| set-parameters: | |
| runs-on: linux.2xlarge | |
| outputs: | |
| models: ${{ steps.set-parameters.outputs.models }} | |
| devices: ${{ steps.set-parameters.outputs.devices }} | |
| delegates: ${{ steps.set-parameters.outputs.delegates }} | |
| steps: | |
| - name: Set parameters | |
| id: set-parameters | |
| shell: bash | |
| env: | |
| # Separate default values from the workflow dispatch. To ensure defaults are accessible | |
| # during scheduled runs and to provide flexibility for different defaults between | |
| # on-demand and periodic benchmarking. | |
| CRON_DEFAULT_MODELS: "stories110M,mv3,mv2,ic4,ic3,resnet50,edsr,mobilebert,w2l" | |
| CRON_DEFAULT_DEVICES: "apple_iphone_15" | |
| CRON_DEFAULT_DELEGATES: "xnnpack,coreml,mps" | |
| run: | | |
| set -ex | |
| MODELS="${{ inputs.models }}" | |
| if [ -z "$MODELS" ]; then | |
| MODELS="$CRON_DEFAULT_MODELS" | |
| fi | |
| DEVICES="${{ inputs.devices }}" | |
| if [ -z "$DEVICES" ]; then | |
| DEVICES="$CRON_DEFAULT_DEVICES" | |
| fi | |
| DELEGATES="${{ inputs.delegates }}" | |
| if [ -z "$DELEGATES" ]; then | |
| DELEGATES="$CRON_DEFAULT_DELEGATES" | |
| fi | |
| # Mapping devices to their corresponding device-pool-arn | |
| declare -A DEVICE_POOL_ARNS | |
| DEVICE_POOL_ARNS[apple_iphone_15]="arn:aws:devicefarm:us-west-2:308535385114:devicepool:02a2cf0f-6d9b-45ee-ba1a-a086587469e6/3b5acd2e-92e2-4778-b651-7726bafe129d" | |
| # Resolve device names with their corresponding ARNs | |
| if [[ ! $(echo "$DEVICES" | jq empty 2>/dev/null) ]]; then | |
| DEVICES=$(echo "$DEVICES" | jq -Rc 'split(",")') | |
| fi | |
| declare -a MAPPED_ARNS=() | |
| for DEVICE in $(echo "$DEVICES" | jq -r '.[]'); do | |
| if [[ -z "${DEVICE_POOL_ARNS[$DEVICE]}" ]]; then | |
| echo "Error: No ARN found for device '$DEVICE'. Abort." >&2 | |
| exit 1 | |
| fi | |
| MAPPED_ARNS+=("${DEVICE_POOL_ARNS[$DEVICE]}") | |
| done | |
| echo "models=$(echo $MODELS | jq -Rc 'split(",")')" >> $GITHUB_OUTPUT | |
| MAPPED_ARNS_JSON=$(printf '%s\n' "${MAPPED_ARNS[@]}" | jq -R . | jq -s .) | |
| echo "devices=$(echo "$MAPPED_ARNS_JSON" | jq -c .)" >> $GITHUB_OUTPUT | |
| echo "delegates=$(echo $DELEGATES | jq -Rc 'split(",")')" >> $GITHUB_OUTPUT | |
| export-models: | |
| name: export-models | |
| uses: pytorch/test-infra/.github/workflows/macos_job.yml@main | |
| needs: set-parameters | |
| strategy: | |
| matrix: | |
| model: ${{ fromJson(needs.set-parameters.outputs.models) }} | |
| delegate: ${{ fromJson(needs.set-parameters.outputs.delegates) }} | |
| fail-fast: false | |
| with: | |
| # NB: Need to use our AWS MacOS runner to upload large models to S3 | |
| runner: macos-m1-stable | |
| python-version: '3.11' | |
| submodules: 'true' | |
| timeout: 60 | |
| upload-artifact: ios-models | |
| upload-artifact-to-s3: true | |
| script: | | |
| set -eux | |
| echo "::group::Setting up CI environment" | |
| .ci/scripts/setup-conda.sh | |
| BUILD_TOOL=cmake | |
| # Setup MacOS dependencies as there is no Docker support on MacOS atm | |
| GITHUB_RUNNER=1 PYTHON_EXECUTABLE=python ${CONDA_RUN} --no-capture-output \ | |
| .ci/scripts/setup-macos.sh "${BUILD_TOOL}" | |
| if [[ ${{ matrix.delegate }} == "coreml" ]]; then | |
| PYTHON_EXECUTABLE=python ${CONDA_RUN} --no-capture-output \ | |
| backends/apple/coreml/scripts/install_requirements.sh | |
| fi | |
| if [[ ${{ matrix.delegate }} == "mps" ]]; then | |
| PYTHON_EXECUTABLE=python ${CONDA_RUN} --no-capture-output \ | |
| backends/apple/mps/install_requirements.sh | |
| fi | |
| ARTIFACTS_DIR_NAME=artifacts-to-be-uploaded/${{ matrix.model }}_${{ matrix.delegate }} | |
| echo "::endgroup::" | |
| echo "::group::Exporting ${{ matrix.delegate }} model: ${{ matrix.model }}" | |
| BUILD_MODE="cmake" | |
| DTYPE="fp32" | |
| if [[ ${{ matrix.model }} =~ ^stories* ]]; then | |
| # Install requirements for export_llama | |
| PYTHON_EXECUTABLE=python ${CONDA_RUN} --no-capture-output \ | |
| bash examples/models/llama/install_requirements.sh | |
| # Test llama2 | |
| if [[ ${{ matrix.delegate }} == "xnnpack" ]]; then | |
| DELEGATE_CONFIG="xnnpack+custom+qe" | |
| elif [[ ${{ matrix.delegate }} == "coreml" ]]; then | |
| DELEGATE_CONFIG="coreml" | |
| elif [[ ${{ matrix.delegate }} == "mps" ]]; then | |
| DELEGATE_CONFIG="mps" | |
| fi | |
| PYTHON_EXECUTABLE=python ${CONDA_RUN} --no-capture-output \ | |
| bash .ci/scripts/test_llama.sh "${{ matrix.model }}" "${BUILD_MODE}" "${DTYPE}" "${DELEGATE_CONFIG}" "${ARTIFACTS_DIR_NAME}" | |
| else | |
| PYTHON_EXECUTABLE=python ${CONDA_RUN} --no-capture-output \ | |
| bash .ci/scripts/test_model.sh "${{ matrix.model }}" "${BUILD_MODE}" "${{ matrix.delegate }}" "${ARTIFACTS_DIR_NAME}" | |
| fi | |
| echo "::endgroup::" | |
| build-benchmark-app: | |
| name: build-benchmark-app | |
| uses: pytorch/test-infra/.github/workflows/macos_job.yml@main | |
| needs: | |
| - set-parameters | |
| secrets: inherit | |
| with: | |
| runner: macos-latest-xlarge | |
| python-version: '3.11' | |
| submodules: 'true' | |
| ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }} | |
| upload-artifact: ios-apps | |
| secrets-env: BUILD_CERTIFICATE_BASE64 EXECUTORCH_BENCHMARK_BUILD_PROVISION_PROFILE_BASE64 KEYCHAIN_PASSWORD | |
| timeout: 90 | |
| script: | | |
| set -eux | |
| echo "::group::Setting up CI environment" | |
| .ci/scripts/setup-conda.sh | |
| BUILD_TOOL=cmake | |
| # Setup MacOS dependencies as there is no Docker support on MacOS atm | |
| GITHUB_RUNNER=1 PYTHON_EXECUTABLE=python ${CONDA_RUN} --no-capture-output \ | |
| .ci/scripts/setup-macos.sh "${BUILD_TOOL}" | |
| export ARTIFACTS_DIR_NAME=artifacts-to-be-uploaded | |
| # Setup Apple certificate for iOS development | |
| BUILD_PROVISION_PROFILE_BASE64="${SECRET_EXECUTORCH_BENCHMARK_BUILD_PROVISION_PROFILE_BASE64}" \ | |
| BUILD_CERTIFICATE_BASE64="${SECRET_BUILD_CERTIFICATE_BASE64}" \ | |
| KEYCHAIN_PASSWORD="${SECRET_KEYCHAIN_PASSWORD}" \ | |
| .ci/scripts/setup-ios.sh | |
| # Install CoreML Backend Requirements | |
| PYTHON_EXECUTABLE=python ${CONDA_RUN} --no-capture-output \ | |
| backends/apple/coreml/scripts/install_requirements.sh | |
| # Install MPS Backend Requirements | |
| PYTHON_EXECUTABLE=python ${CONDA_RUN} --no-capture-output \ | |
| backends/apple/mps/install_requirements.sh | |
| echo "::endgroup::" | |
| echo "::group::Build ExecuTorch iOS frameworks" | |
| FRAMEWORKS=( | |
| "executorch" | |
| "backend_coreml" | |
| "backend_mps" | |
| "backend_xnnpack" | |
| "kernels_custom" | |
| "kernels_optimized" | |
| "kernels_portable" | |
| "kernels_quantized" | |
| ) | |
| # Build Release iOS Frameworks | |
| PYTHON_EXECUTABLE=python ${CONDA_RUN} --no-capture-output \ | |
| build/build_apple_frameworks.sh --coreml --custom --mps --optimized --portable --quantized --xnnpack | |
| mkdir -p extension/benchmark/apple/Benchmark/Frameworks | |
| for FRAMEWORK in "${FRAMEWORKS[@]}"; do ( | |
| cp -r "cmake-out/${FRAMEWORK}.xcframework" extension/benchmark/apple/Benchmark/Frameworks/ | |
| ) done | |
| echo "::endgroup::" | |
| # NB: Although exported models can be copied to this directory and bundled together with the | |
| # app, we don't use this in CI and rely on AWS extra data parameter to make the model and the | |
| # tokenizer available to the benchmark. This decouples the app and the model. We just need to | |
| # create the directory here to pass the build | |
| mkdir -p extension/benchmark/apple/Benchmark/Models | |
| ${CONDA_RUN} --no-capture-output \ | |
| build/build_apple_llm_demo.sh ${ARTIFACTS_DIR_NAME} | |
| upload-benchmark-app: | |
| needs: build-benchmark-app | |
| runs-on: linux.2xlarge | |
| steps: | |
| - name: Download the apps from GitHub | |
| uses: actions/download-artifact@v3 | |
| with: | |
| # The name here needs to match the name of the upload-artifact parameter | |
| name: ios-apps | |
| path: ${{ runner.temp }}/artifacts/ | |
| - name: Verify the apps | |
| shell: bash | |
| working-directory: ${{ runner.temp }}/artifacts/ | |
| run: | | |
| ls -lah ./ | |
| - name: Upload the apps to S3 | |
| uses: seemethere/upload-artifact-s3@v5 | |
| with: | |
| s3-bucket: gha-artifacts | |
| s3-prefix: | | |
| ${{ github.repository }}/${{ github.run_id }}/artifacts | |
| retention-days: 14 | |
| if-no-files-found: ignore | |
| path: ${{ runner.temp }}/artifacts/ | |
| benchmark-on-device: | |
| if: always() | |
| needs: | |
| - set-parameters | |
| - upload-benchmark-app | |
| - export-models | |
| permissions: | |
| id-token: write | |
| contents: read | |
| uses: pytorch/test-infra/.github/workflows/mobile_job.yml@main | |
| strategy: | |
| matrix: | |
| model: ${{ fromJson(needs.set-parameters.outputs.models) }} | |
| delegate: ${{ fromJson(needs.set-parameters.outputs.delegates) }} | |
| device: ${{ fromJson(needs.set-parameters.outputs.devices) }} | |
| fail-fast: false | |
| with: | |
| # Due to scheduling a job may be pushed beyond the default 60m threshold | |
| timeout: 120 | |
| device-type: ios | |
| # For iOS testing, the runner just needs to call AWS Device Farm, so there is no need to run this on macOS | |
| runner: linux.2xlarge | |
| test-infra-ref: '' | |
| # This is the ARN of ExecuTorch project on AWS | |
| project-arn: arn:aws:devicefarm:us-west-2:308535385114:project:02a2cf0f-6d9b-45ee-ba1a-a086587469e6 | |
| device-pool-arn: ${{ matrix.device }} | |
| # Uploaded to S3 from the previous job | |
| ios-ipa-archive: https://gha-artifacts.s3.amazonaws.com/${{ github.repository }}/${{ github.run_id }}/artifacts/Benchmark.ipa | |
| ios-xctestrun-zip: https://gha-artifacts.s3.amazonaws.com/${{ github.repository }}/${{ github.run_id }}/artifacts/Benchmark.xctestrun.zip | |
| test-spec: ${{ inputs.test_spec || 'https://ossci-ios.s3.amazonaws.com/executorch/default-ios-device-farm-appium-test-spec.yml' }} | |
| extra-data: https://gha-artifacts.s3.amazonaws.com/${{ github.repository }}/${{ github.run_id }}/artifacts/${{ matrix.model }}_${{ matrix.delegate }}/model.zip | |
| upload-benchmark-results: | |
| needs: | |
| - benchmark-on-device | |
| if: always() | |
| runs-on: linux.2xlarge | |
| environment: upload-benchmark-results | |
| permissions: | |
| id-token: write | |
| contents: read | |
| steps: | |
| - uses: actions/checkout@v3 | |
| with: | |
| submodules: false | |
| - name: Authenticate with AWS | |
| uses: aws-actions/configure-aws-credentials@v4 | |
| with: | |
| role-to-assume: arn:aws:iam::308535385114:role/gha_workflow_upload-benchmark-results | |
| # The max duration enforced by the server side | |
| role-duration-seconds: 18000 | |
| aws-region: us-east-1 | |
| - name: Setup conda | |
| uses: pytorch/test-infra/.github/actions/setup-miniconda@main | |
| with: | |
| python-version: '3.10' | |
| - name: Download the list of artifacts from S3 | |
| env: | |
| ARTIFACTS_S3_DIR: s3://gha-artifacts/device_farm/${{ github.run_id }}/${{ github.run_attempt }}/artifacts/ | |
| shell: bash | |
| run: | | |
| set -eux | |
| ${CONDA_RUN} python -mpip install awscli==1.32.18 | |
| mkdir -p artifacts | |
| pushd artifacts | |
| ${CONDA_RUN} aws s3 sync "${ARTIFACTS_S3_DIR}" . | |
| popd | |
| ls -lah artifacts | |
| - name: Extract the benchmark results JSON | |
| shell: bash | |
| run: | | |
| set -eux | |
| mkdir -p benchmark-results | |
| for ARTIFACTS_BY_JOB in artifacts/*.json; do | |
| [ -f "${ARTIFACTS_BY_JOB}" ] || break | |
| echo "${ARTIFACTS_BY_JOB}" | |
| ${CONDA_RUN} python .github/scripts/extract_benchmark_results.py \ | |
| --artifacts "${ARTIFACTS_BY_JOB}" \ | |
| --output-dir benchmark-results \ | |
| --repo ${{ github.repository }} \ | |
| --head-branch ${{ github.head_ref || github.ref_name }} \ | |
| --workflow-name "${{ github.workflow }}" \ | |
| --workflow-run-id ${{ github.run_id }} \ | |
| --workflow-run-attempt ${{ github.run_attempt }} | |
| done | |
| ls -lah benchmark-results | |
| for BENCHMARK_RESULTS in benchmark-results/*.json; do | |
| cat "${BENCHMARK_RESULTS}" | |
| echo | |
| done | |
| - name: Upload the benchmark results | |
| uses: pytorch/test-infra/.github/actions/upload-benchmark-results@main | |
| with: | |
| benchmark-results-dir: 'benchmark-results' | |
| dry-run: false |