main.yml

name: CI

permissions:
  contents: read

on:
  push:
    branches: [ "main" ]
    tags:
      - v*
  pull_request:

concurrency:
  group: ${{ github.workflow }}-${{ github.ref }}
  cancel-in-progress: true

jobs:
  lint:
    name: Lint
    runs-on: ubuntu-latest
    steps:
    - uses: actions/checkout@v7
    - name: Set up Python "3.13"
      uses: actions/setup-python@v7
      with:
        python-version: "3.13"
    - name: Lint with lintrunner
      uses: justinchuby/lintrunner-action@main

  detect-affected:
    name: Detect affected models
    runs-on: ubuntu-latest
    outputs:
      affected: ${{ steps.detect.outputs.affected }}
      golden_l4_cases: ${{ steps.detect.outputs.golden_l4_cases }}
      golden_l5_cases: ${{ steps.detect.outputs.golden_l5_cases }}
      run_all: ${{ steps.detect.outputs.run_all }}
      has_affected: ${{ steps.detect.outputs.has_affected }}
      has_l4_affected: ${{ steps.detect.outputs.has_l4_affected }}
      has_l5_affected: ${{ steps.detect.outputs.has_l5_affected }}
    steps:
      - uses: actions/checkout@v7
        with:
          fetch-depth: 0
      - name: Get changed files
        id: changed
        env:
          BASE_REF: ${{ github.base_ref }}
          BEFORE: ${{ github.event.before }}
          SHA: ${{ github.sha }}
        run: |
          if [ "${{ github.event_name }}" = "pull_request" ]; then
            git diff --name-only origin/"$BASE_REF"...HEAD > changed_files.txt
          elif [ "${{ github.event_name }}" = "push" ] && \
               [ -n "$BEFORE" ] && \
               [ "$BEFORE" != "0000000000000000000000000000000000000000" ]; then
            # Push with a known previous commit: use the full push range so
            # multi-commit pushes don't silently miss changed files.
            git diff --name-only "$BEFORE"..."$SHA" > changed_files.txt
          else
            # workflow_dispatch, new-branch push (before=0000...), or unknown
            # event: fall back to HEAD~1 when it exists, or diff-tree for the
            # root commit of a new repository.
            if git rev-parse --verify HEAD~1 >/dev/null 2>&1; then
              git diff --name-only HEAD~1...HEAD > changed_files.txt
            else
              git diff-tree --root --no-commit-id --name-only -r HEAD > changed_files.txt
            fi
          fi
          echo "Changed files:"
          cat changed_files.txt
      - name: Detect affected models
        id: detect
        run: |
          output=$(python scripts/detect_affected_models.py \
            --stdin --output-format github < changed_files.txt)
          echo "$output"
          echo "$output" >> "$GITHUB_OUTPUT"

  smoke-test:
    name: L1 Smoke Test
    runs-on: ubuntu-latest
    needs: [detect-affected]
    if: always() && !cancelled()
    steps:
      - uses: actions/checkout@v7
      - name: Setup Python
        uses: actions/setup-python@v7
        with:
          python-version: "3.12"
      - name: Cache pip packages
        uses: actions/cache@v6
        with:
          path: ~/.cache/pip
          key: pip-smoke-${{ hashFiles('pyproject.toml', 'requirements/ci/requirements.txt') }}
          restore-keys: |
            pip-smoke-
      - name: Install PyTorch CPU
        run: pip install torch --index-url https://download.pytorch.org/whl/cpu
      - name: Install dependencies
        run: |
          pip install -r requirements/ci/requirements.txt
          pip install onnxruntime
          pip install -e '.[testing]'
      - name: Run L1 smoke tests (all model configs)
        run: |
          pytest tests/build_graph -v --tb=short -n auto \
            --cov=src --cov-report=xml --cov-branch --junitxml junit.xml
        timeout-minutes: 10
      - name: Upload coverage to Codecov
        if: always()
        uses: codecov/codecov-action@v7
        with:
          token: ${{ secrets.CODECOV_TOKEN }}
          flags: smoke
      - name: Upload test results to Codecov
        if: ${{ !cancelled() }}
        uses: codecov/codecov-action@v7
        with:
          token: ${{ secrets.CODECOV_TOKEN }}
          flags: smoke
          report-type: test_results

  synthetic-parity:
    name: L3 Synthetic Parity
    runs-on: ubuntu-latest
    needs: [detect-affected]
    if: >-
      always() && !cancelled() &&
      (needs.detect-affected.result == 'failure' ||
       needs.detect-affected.outputs.has_l4_affected == 'true')
    steps:
      - uses: actions/checkout@v7
      - name: Setup Python
        uses: actions/setup-python@v7
        with:
          python-version: "3.12"
      - name: Cache pip packages
        uses: actions/cache@v6
        with:
          path: ~/.cache/pip
          key: pip-parity-${{ hashFiles('pyproject.toml', 'requirements/ci/requirements.txt') }}
          restore-keys: |
            pip-parity-
      - name: Install PyTorch CPU
        run: pip install torch --index-url https://download.pytorch.org/whl/cpu
      - name: Install dependencies
        run: |
          pip install -r requirements/ci/requirements.txt
          pip install onnxruntime
          pip install -e '.[testing,transformers]'
          pip install --force-reinstall --no-deps -r requirements/ci/vibevoice.txt
      - name: Run L3 synthetic parity tests
        env:
          HF_TOKEN: ${{ secrets.HF_TOKEN }}
        run: |
          if [ ! -f tests/synthetic_parity_test.py ]; then
            echo "tests/synthetic_parity_test.py not found — skipping"
            exit 0
          fi
          if [ "${{ needs.detect-affected.result }}" = "failure" ] || \
             [ "${{ needs.detect-affected.outputs.run_all }}" = "true" ]; then
            echo "Running all synthetic parity tests"
            pytest tests/synthetic_parity_test.py -v --tb=short -n auto \
              --cov=src --cov-report=xml --cov-branch --junitxml junit.xml
          else
            AFFECTED='${{ needs.detect-affected.outputs.affected }}'
            echo "Running parity for affected models: $AFFECTED"
            MODELS=$(echo "$AFFECTED" | python -c "import json, sys; print(','.join(json.load(sys.stdin)))")
            if [ -n "$MODELS" ]; then
              pytest tests/synthetic_parity_test.py -v --tb=short -n auto --models "$MODELS" \
                --cov=src --cov-report=xml --cov-branch --junitxml junit.xml
            else
              echo "No affected models — skipping"
            fi
          fi
        timeout-minutes: 20
      - name: Upload coverage to Codecov
        if: always()
        uses: codecov/codecov-action@v7
        with:
          token: ${{ secrets.CODECOV_TOKEN }}
          flags: synthetic-parity
      - name: Upload test results to Codecov
        if: ${{ !cancelled() }}
        uses: codecov/codecov-action@v7
        with:
          token: ${{ secrets.CODECOV_TOKEN }}
          flags: synthetic-parity
          report-type: test_results

  golden-comparison:
    name: L4 Golden Comparison
    needs: [detect-affected]
    if: >-
      always() && !cancelled() &&
      (needs.detect-affected.result == 'failure' ||
       needs.detect-affected.outputs.has_l4_affected == 'true')
    uses: ./.github/workflows/gpu_l4_golden_parity.yml
    with:
      affected_models: ${{ needs.detect-affected.outputs.affected }}
      affected_cases: ${{ needs.detect-affected.outputs.golden_l4_cases }}
    secrets:
      HF_TOKEN: ${{ secrets.HF_TOKEN }}
      CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}

  generation-e2e:
    name: L5 Generation E2E
    needs: [detect-affected]
    if: >-
      always() && !cancelled() &&
      (needs.detect-affected.result == 'failure' ||
       needs.detect-affected.outputs.has_l5_affected == 'true')
    uses: ./.github/workflows/gpu_l5_generation_e2e.yml
    with:
      affected_models: ${{ needs.detect-affected.outputs.affected }}
      affected_cases: ${{ needs.detect-affected.outputs.golden_l5_cases }}
    secrets:
      HF_TOKEN: ${{ secrets.HF_TOKEN }}
      CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}

  ort-genai-e2e:
    name: ORT GenAI E2E
    needs: [detect-affected]
    if: >-
      always() && !cancelled() &&
      (needs.detect-affected.result == 'failure' ||
       needs.detect-affected.outputs.has_affected == 'true')
    uses: ./.github/workflows/ort_genai_e2e.yml
    with:
      run_real: false

  test:
    name: test
    strategy:
      fail-fast: false
      matrix:
        os: [ubuntu-latest, windows-latest]
        python_version: ["3.11", "3.12", "3.13"]
    runs-on: ${{ matrix.os }}
    steps:
      - uses: actions/checkout@v7
      - name: Setup Python
        uses: actions/setup-python@v7
        with:
          python-version: ${{ matrix.python_version }}
      - name: Cache pip packages
        uses: actions/cache@v6
        with:
          path: ~/.cache/pip
          key: pip-${{ runner.os }}-${{ matrix.python_version }}-${{ hashFiles('pyproject.toml', 'requirements/ci/requirements.txt') }}
          restore-keys: |
            pip-${{ runner.os }}-${{ matrix.python_version }}-
      - name: Install PyTorch CPU
        run: pip install torch --index-url https://download.pytorch.org/whl/cpu
      - name: Install dependencies
        run: |
          pip install -r requirements/ci/requirements.txt
          pip install onnxruntime
          pip install -e '.[testing]'
          pip install --force-reinstall --no-deps -r requirements/ci/vibevoice.txt
      - name: Run tests
        run: |
          pytest -n auto -m 'not integration and not arch_validation' --cov=src --cov-report=xml --cov-branch --junitxml junit.xml
      - name: Upload coverage to Codecov
        if: always()
        uses: codecov/codecov-action@v7
        with:
          token: ${{ secrets.CODECOV_TOKEN }}
          flags: unittests
      - name: Upload test results to Codecov
        if: ${{ !cancelled() }}
        uses: codecov/codecov-action@v7
        with:
          token: ${{ secrets.CODECOV_TOKEN }}
          flags: unittests
          report-type: test_results

  integration-fast:
    name: Integration (fast)
    runs-on:
      - self-hosted
      - "1ES.Pool=onnxruntime-ep-mobius-github-linux-a10"
      - "JobId=gpu-integration-fast-${{ github.run_id }}-${{ github.run_number }}-${{ github.run_attempt }}"
    steps:
      - uses: actions/checkout@v7
      - name: Setup Python
        uses: actions/setup-python@v7
        with:
          python-version: "3.12"
      - name: Cache HuggingFace models
        uses: actions/cache@v6
        with:
          path: ~/.cache/huggingface
          key: hf-integration-fast-${{ hashFiles('tests/integration/_support.py', 'tests/integration/text_test.py', 'tests/integration/generation_test.py', 'tests/integration/encoder_media_seq2seq_test.py', 'tests/integration/architectures_test.py', 'tests/integration/qwen3_5_vl_test.py', 'tests/gguf_small_model_runtime_integration_test.py') }}
          restore-keys: |
            hf-integration-fast-
      - name: Install PyTorch CUDA
        run: pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128
      - name: Install dependencies
        run: |
          pip install -r requirements/ci/requirements.txt
          pip install soundfile librosa ml_dtypes flatbuffers numpy packaging protobuf sympy coloredlogs
          pip install -e '.[testing]'
          # PyPI's ORT 1.27+ wheel requires CUDA 13, while these runners use CUDA 12.
          pip install --pre --no-deps --index-url https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/onnxruntime-cuda-12/pypi/simple/ onnxruntime-gpu
          pip install --no-deps onnxruntime-genai-cuda
      - name: Run fast integration tests
        env:
          HF_TOKEN: ${{ secrets.HF_TOKEN }}
        run: |
          pytest tests/integration/text_test.py tests/integration/generation_test.py \
            tests/integration/encoder_media_seq2seq_test.py \
            tests/integration/architectures_test.py \
            tests/integration/qwen3_5_vl_test.py \
            tests/gguf_small_model_runtime_integration_test.py -m integration_fast -v \
            -k "smollm-135m or smollm2-135m-instruct-f16 or albert-base or t5-small or dinov2-small or wav2vec2-base or whisper-tiny or test_gemma3_multimodal or test_qwen35 or test_qwen3_next or test_deepseek or test_sam_vit or test_ocr2" \
            --cov=src --cov-report=xml --cov-branch --junitxml junit.xml
        timeout-minutes: 15
      - name: Upload coverage to Codecov
        if: always()
        uses: codecov/codecov-action@v7
        with:
          token: ${{ secrets.CODECOV_TOKEN }}
          flags: integration
      - name: Upload test results to Codecov
        if: ${{ !cancelled() }}
        uses: codecov/codecov-action@v7
        with:
          token: ${{ secrets.CODECOV_TOKEN }}
          flags: integration
          report-type: test_results

  build:
    name: Build
    needs: [test]
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v7
        with:
          persist-credentials: false
      - name: Setup Python
        uses: actions/setup-python@v7
        with:
          python-version: "3.12"
      - name: Install Python build dependencies
        run: |
          pip install build wheel
      - name: Build wheels
        run: |
          python -m build