benchmarks_pr.yml

name: Continuous Benchmark (PR)

on:
  pull_request:

permissions:
  contents: read
  issues: write
  pull-requests: write

concurrency:
  # Documentation suggests ${{ github.head_ref }}, but that's only available on pull_request/pull_request_target triggers, so using ${{ github.ref }}.
  # On master, we want all builds to complete even if merging happens faster to make it easier to discover at which point something broke.
  group: ${{ github.workflow }}-${{ github.ref == 'refs/heads/main' && format('ci-master-{0}', github.sha) || format('ci-{0}', github.ref) }}
  cancel-in-progress: true

jobs:
  benchmark:
    name: ${{ matrix.device }} Pytest benchmark
    runs-on: linux.g5.4xlarge.nvidia.gpu
    strategy:
      matrix:
        device: [CPU, GPU]
        python-version: ['3.12']
    defaults:
      run:
        shell: bash -l {0}
    container:
      image: nvidia/cuda:12.8.0-runtime-ubuntu22.04
      options: --gpus all
    steps:
      - name: Who triggered this?
        run: |
          echo "Action triggered by ${{ github.event.pull_request.html_url }}"
      - name: Install deps
        run: |
          export TZ=Europe/London
          export DEBIAN_FRONTEND=noninteractive  # tzdata bug
          apt-get update -y
          apt-get update -y
          apt-get upgrade -y
          apt-get -y install gcc curl g++ unzip wget sudo git cmake
      - name: Check ldd --version
        run: ldd --version
      - name: Checkout
        uses: actions/checkout@v4
        with:
          fetch-depth: 50 # this is to make sure we obtain the target base commit
      - name: Python Setup
        uses: actions/setup-python@v5
        with:
          python-version: '3.12'

      - name: Setup git
        run: git config --global --add safe.directory /__w/tensordict/tensordict
      - name: setup Path
        run: |
          echo /usr/local/bin >> $GITHUB_PATH
      - name: Setup Environment
        run: |
          set -e
          set -x
          curl -LsSf https://astral.sh/uv/install.sh | sh
          echo "$HOME/.local/bin" >> $GITHUB_PATH
          source $HOME/.local/bin/env
          # Create a local venv for the benchmark run
          uv venv .venv/local --python ${{ matrix.python-version }}
          source .venv/local/bin/activate
          echo "=== uv version ==="
          uv --version
          uv pip install --pre torch --index-url https://download.pytorch.org/whl/nightly/cu128
          uv pip install "pybind11[global]" "setuptools" "wheel" "ninja"
          uv pip install pytest pytest-benchmark "zarr>=3.0"
          # Do not resolve runtime dependencies here: we want to keep the PyTorch build (nightly)
          # that was explicitly installed above, and avoid any solver replacing it.
          uv pip install -e . --no-deps
          uv run --active python -c "import torch; print(torch.__version__)"
          ${{ matrix.device == 'CPU' && 'export CUDA_VISIBLE_DEVICES=' || '' }}
      - name: check GPU presence
        if: matrix.device == 'GPU'
        run: |
          source .venv/local/bin/activate
          uv run --active python -c """import torch
          print(f'CUDA available: {torch.cuda.is_available()}')
          print(f'CUDA device count: {torch.cuda.device_count()}')
          print(f'CUDA version: {torch.version.cuda}')
          assert torch.cuda.is_available() and torch.cuda.device_count() > 0, 'CUDA is not available or no GPU devices found'
          """
      - name: Setup benchmarks
        run: |
          source .venv/local/bin/activate
          echo "BASE_SHA=$(echo ${{ github.event.pull_request.base.sha }} | cut -c1-8)" >> $GITHUB_ENV
          echo "HEAD_SHA=$(echo ${{ github.event.pull_request.head.sha }} | cut -c1-8)" >> $GITHUB_ENV
          echo "BASELINE_JSON=${RUNNER_TEMP}/${{ matrix.device }}-baseline.json" >> $GITHUB_ENV
          echo "CONTENDER_JSON=${RUNNER_TEMP}/${{ matrix.device }}-contender.json" >> $GITHUB_ENV
          echo "PR_COMMENT=$(mktemp)" >>  $GITHUB_ENV
      - name: Run benchmarks
        run: |
          set -e
          set -x
          source .venv/local/bin/activate
          cd benchmarks/
          # Check Python version to ensure compatibility with PyTorch Dynamo
          uv run --active python --version
          uv run --active python -c "import sys; actual_version = f'{sys.version_info.major}.{sys.version_info.minor}'; expected_version = '${{ matrix.python-version }}'; print(f'Expected: {expected_version}, Actual: {actual_version}'); exit(1) if actual_version != expected_version else None; exit(1) if sys.version_info >= (3, 13) else print(f'Python version check passed: {actual_version}')"
          export TORCHDYNAMO_INLINE_INBUILT_NN_MODULES=1
          export TD_GET_DEFAULTS_TO_NONE=1
          RUN_BENCHMARK="uv run --active pytest -vvv --rank 0 --benchmark-json "
          git checkout ${{ github.event.pull_request.base.sha }}
          $RUN_BENCHMARK ${{ env.BASELINE_JSON }}
          git checkout ${{ github.event.pull_request.head.sha }}
          $RUN_BENCHMARK ${{ env.CONTENDER_JSON }}
      - name: Align benchmark JSONs
        run: |
          python3 -c "
          import json
          with open('${{ env.BASELINE_JSON }}') as f:
              baseline = json.load(f)
          with open('${{ env.CONTENDER_JSON }}') as f:
              contender = json.load(f)
          base_names = {b['fullname'] for b in baseline['benchmarks']}
          cont_names = {b['fullname'] for b in contender['benchmarks']}
          common = base_names & cont_names
          new_tests = cont_names - base_names
          removed_tests = base_names - cont_names
          if new_tests:
              print(f'New benchmarks (skipped in comparison): {new_tests}')
          if removed_tests:
              print(f'Removed benchmarks (skipped in comparison): {removed_tests}')
          baseline['benchmarks'] = [b for b in baseline['benchmarks'] if b['fullname'] in common]
          contender['benchmarks'] = [b for b in contender['benchmarks'] if b['fullname'] in common]
          with open('${{ env.BASELINE_JSON }}', 'w') as f:
              json.dump(baseline, f)
          with open('${{ env.CONTENDER_JSON }}', 'w') as f:
              json.dump(contender, f)
          "
      - name: Upload benchmark comparison data
        uses: actions/upload-artifact@v4
        with:
          name: ${{ matrix.device }}-benchmark-comparison-data
          path: |
            ${{ env.BASELINE_JSON }}
            ${{ env.CONTENDER_JSON }}
      - name: Publish results
        continue-on-error: true
        uses: apbard/pytest-benchmark-commenter@v3
        with:
          token: ${{ secrets.GITHUB_TOKEN }}
          benchmark-file: ${{ env.CONTENDER_JSON }}
          comparison-benchmark-file: ${{ env.BASELINE_JSON }}
          benchmark-metrics: 'name,max,mean,ops'
          comparison-benchmark-metric: 'ops'
          comparison-higher-is-better: true
          comparison-threshold: 5
          benchmark-title: 'Result of ${{ matrix.device }} Benchmark Tests'