Create a workflow to run benchmarks #41
Workflow file for this run
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: Benchmarks | |
| on: | |
| pull_request: | |
| branches: | |
| - main | |
| workflow_dispatch: | |
| inputs: | |
| halt-for-connection: | |
| description: 'Should this workflow run wait for a remote connection?' | |
| type: choice | |
| required: true | |
| default: 'no' | |
| options: | |
| - 'yes' | |
| - 'no' | |
| jobs: | |
| # jax-build: | |
| # strategy: | |
| # matrix: | |
| # runner: ["linux-x86-g2-48-l4-4gpu"] | |
| # runs-on: ${{ matrix.runner }} | |
| # container: | |
| # image: "gcr.io/tensorflow-testing/nosla-cuda12.3-cudnn9.1-ubuntu20.04-manylinux2014-multipython:latest" | |
| # env: | |
| # JAXCI_HERMETIC_PYTHON_VERSION: 3.11 | |
| # steps: | |
| # - name: Checkout JAX Fork | |
| # uses: actions/checkout@v3 | |
| # with: | |
| # repository: 'google-ml-infra/jax-fork' | |
| # path: jax-fork | |
| # - name: Install JAX Dependencies | |
| # working-directory: jax-fork | |
| # run: | | |
| # python -m pip install --upgrade pip | |
| # pip install pytest | |
| # pip install absl-py | |
| # pip install "jax[cuda12_pip]" # Adjust CUDA version if needed | |
| # pip install google-benchmark | |
| # - name: Run JAX Multiprocess GPU Test | |
| # working-directory: jax-fork | |
| # continue-on-error: true | |
| # run: python -m pytest tests/multiprocess_gpu_test.py | |
| # - name: Build XLA GPU Atomic Test | |
| # working-directory: xla | |
| # continue-on-error: true | |
| # run: bazel build -c opt --config=cuda //xla/service/gpu/tests:gpu_atomic_test | |
| # - name: Run XLA GPU Atomic Test | |
| # working-directory: xla | |
| # continue-on-error: true | |
| # run: bazel test -c opt --config=cuda //xla/service/gpu/tests:gpu_atomic_test | |
| build-xla-gpu: | |
| runs-on: linux-x86-g2-48-l4-4gpu # Use a GPU-enabled runner | |
| container: | |
| image: "gcr.io/tensorflow-testing/nosla-cuda12.3-cudnn9.1-ubuntu20.04-manylinux2014-multipython:latest" | |
| options: --gpus all --privileged # Might need privileged mode, use with caution | |
| steps: | |
| - name: Checkout XLA | |
| uses: actions/checkout@v3 | |
| with: | |
| repository: openxla/xla # Replace with your fork if needed | |
| path: xla | |
| # - name: Build XLA with GPU support | |
| # working-directory: xla | |
| # continue-on-error: true | |
| # run: bazel build --build_tag_filters=-no_oss,requires-gpu-nvidia,gpu,-rocm-only --test_tag_filters=-no_oss,requires-gpu-nvidia,gpu,-rocm-only,requires-gpu-sm75-only,requires-gpu-sm60,requires-gpu-sm70,-requires-gpu-sm80,-requires-gpu-sm80-only,-requires-gpu-sm90,-requires-gpu-sm90-only,-requires-gpu-sm100,-requires-gpu-sm100-only,-requires-gpu-amd --config=warnings --config=rbe_linux_cuda_nvcc --run_under=//tools/ci_build/gpu_build:parallel_gpu_execute --repo_env=TF_CUDA_COMPUTE_CAPABILITIES=7.5 --@cuda_driver//:enable_forward_compatibility=true --test_output=errors --verbose_failures --keep_going --nobuild_tests_only --profile=profile.json.gz --flaky_test_attempts=3 --jobs=150 --bes_upload_mode=fully_async --nobuild -- //xla/... //build_tools/... @tsl//tsl/... | |
| # - name: Run XLA tests with GPU | |
| # working-directory: xla | |
| # continue-on-error: true | |
| # run: bazel test --build_tag_filters=-no_oss,requires-gpu-nvidia,gpu,-rocm-only --test_tag_filters=-no_oss,requires-gpu-nvidia,gpu,-rocm-only,requires-gpu-sm75-only,requires-gpu-sm60,requires-gpu-sm70,-requires-gpu-sm80,-requires-gpu-sm80-only,-requires-gpu-sm90,-requires-gpu-sm90-only,-requires-gpu-sm100,-requires-gpu-sm100-only,-requires-gpu-amd --config=warnings --config=rbe_linux_cuda_nvcc --run_under=//tools/ci_build/gpu_build:parallel_gpu_execute --repo_env=TF_CUDA_COMPUTE_CAPABILITIES=7.5 --@cuda_driver//:enable_forward_compatibility=true --test_output=errors --verbose_failures --keep_going --nobuild_tests_only --profile=profile.json.gz --flaky_test_attempts=3 --jobs=150 --bes_upload_mode=fully_async -- //xla/... //build_tools/... @tsl//tsl/... | |
| # - name: Run Profile Analysis | |
| # working-directory: xla | |
| # continue-on-error: true | |
| # run: bazel analyze-profile profile.json.gz | |
| # - name: Get GPU spec | |
| # working-directory: xla | |
| # continue-on-error: true | |
| # run: nvidia-smi | |
| - name: Wait For Connection | |
| uses: google-ml-infra/actions/ci_connection@main | |
| with: | |
| halt-dispatch-input: ${{ inputs.halt-for-connection }} | |
| - name: Configure XLA | |
| working-directory: xla | |
| run: ./configure.py --backend CUDA --nccl | |
| - name: Set TF_CPP_MAX_VLOG_LEVEL | |
| working-directory: xla | |
| run: echo "TF_CPP_MAX_VLOG_LEVEL=1" >> $GITHUB_ENV | |
| - name: Build hlo_runner_main | |
| working-directory: xla | |
| run: bazel build -c opt --config=cuda --dynamic_mode=off //xla/tools/multihost_hlo_runner:hlo_runner_main | |
| - name: Run XLA GPU microbenchmarks with hlo_runner_main | |
| working-directory: xla | |
| continue-on-error: true | |
| run: | | |
| for file in xla/tools/multihost_hlo_runner/data/*.hlo; do | |
| filename=$(basename "$file") | |
| echo "Running benchmark on $file" | |
| bazel-bin/xla/tools/multihost_hlo_runner/hlo_runner_main "$file" | |
| done |