win-cpu-arm64-build.yml

name: "Windows CPU arm64 Build"
on:
  workflow_dispatch:
  push:
    branches:
      - main
      - rel-*
  pull_request:

concurrency:
  group: ${{ github.workflow }}-${{ github.event_name == 'pull_request' && github.ref || github.sha }}
  cancel-in-progress: true
env:
  binaryDir: 'build/cpu/win-arm64'
  ORT_NIGHTLY_REST_API: "https://feeds.dev.azure.com/aiinfra/PublicPackages/_apis/packaging/Feeds/ORT-Nightly/packages?packageNameQuery=Microsoft.ML.OnnxRuntime&api-version=6.0-preview.1"
  ORT_PACKAGE_NAME: "Microsoft.ML.OnnxRuntime"

jobs:
  windows-cpu-arm64-build:
    runs-on: ["self-hosted", "1ES.Pool=onnxruntime-genai-win11-arm64-cpu2"]
    steps:
      - name: Checkout OnnxRuntime GenAI repo
        uses: actions/checkout@v5
        with:
          submodules: true

      - uses: nuget/setup-nuget@v2
        with:
          nuget-version: '5.x'

      - name: Set up Python 3.12 for ARM64
        uses: actions/setup-python@v6
        with:
          python-version: '3.12'
          architecture: 'arm64'
          cache: 'pip'

      - name: Setup Java 21
        uses: actions/setup-java@v5
        with:
          java-version: '21'
          distribution: 'temurin'
          architecture: 'aarch64'
          cache: 'gradle'

      - name: Setup Gradle
        uses: gradle/actions/setup-gradle@v3
        with:
          gradle-version: '8.6'

      - name: Download OnnxRuntime Nightly
        shell: powershell
        run: |
          $resp = Invoke-RestMethod "${{ env.ORT_NIGHTLY_REST_API }}"
          $ORT_NIGHTLY_VERSION = $resp.value[0].versions[0].normalizedVersion
          Write-Host "$ORT_NIGHTLY_VERSION"
          "ORT_NIGHTLY_VERSION=$ORT_NIGHTLY_VERSION" | Out-File -FilePath $env:GITHUB_ENV -Append
          nuget install ${{ env.ORT_PACKAGE_NAME }} -version $ORT_NIGHTLY_VERSION -x -NonInteractive

      - run: Get-ChildItem  ${{ env.ORT_PACKAGE_NAME }} -Recurse
        continue-on-error: true

      - name: Extract OnnxRuntime library and header files
        run: |
          mkdir ort/lib
          move ${{ env.ORT_PACKAGE_NAME }}/build/native/include ort/
          move ${{ env.ORT_PACKAGE_NAME }}/runtimes/win-arm64/native/* ort/lib/

      - name: Install Rust Toolchain
        run: |
          $exePath = "$env:TEMP\rustup-init.exe"
          (New-Object Net.WebClient).DownloadFile('https://static.rust-lang.org/rustup/dist/aarch64-pc-windows-msvc/rustup-init.exe', $exePath)
          & $exePath -y --default-toolchain=1.86.0
          Add-Content $env:GITHUB_PATH "$env:USERPROFILE\.cargo\bin"

      - name: Install LLVM
        run: |
          choco install llvm --yes
          Add-Content $env:GITHUB_PATH "C:\Program Files\LLVM\bin"

      - name: Configure CMake
        run: |
          python -m pip install wheel requests

          cmake --preset windows_arm64_cpu_release -DENABLE_JAVA=ON -DUSE_GUIDANCE=ON

      - name: Build with CMake
        run: |
          cmake --build --preset windows_arm64_cpu_release --parallel
          cmake --build --preset windows_arm64_cpu_release --target PyPackageBuild

      - name: Build the C Examples
        run: |
          cmake -S examples\c -B examples\c\build `
            -DMODEL_QA=ON -DMODEL_CHAT=ON -DMODEL_MM=ON -DWHISPER= -DNEMOTRON_SPEECH=ON `
            "-DORT_INCLUDE_DIR=$env:GITHUB_WORKSPACE\$env:binaryDir\_deps\ortlib-src\build\native\include" `
            "-DORT_LIB_DIR=$env:GITHUB_WORKSPACE\$env:binaryDir\_deps\ortlib-src\runtimes\win-arm64\native" `
            "-DOGA_INCLUDE_DIR=$env:GITHUB_WORKSPACE\src" `
            "-DOGA_LIB_DIR=$env:GITHUB_WORKSPACE\$env:binaryDir\Release"
          cmake --build examples\c\build --config Release

      - name: Install the Python Wheel and Test Dependencies
        run: |
          # Uninstalling LLVM/Clang as it is no longer required and causes issues with numpy installation
          choco uninstall llvm --yes
          python -m pip install -r test\python\requirements.txt --user
          python -m pip install -r test\python\cpu\torch\requirements.txt --user
          python -m pip install -r test\python\cpu\ort\requirements.txt --user
          python -m pip install (Get-ChildItem ("$env:binaryDir\wheel\*.whl"))

      - name: Run the Java tests
        run: |
          ctest --test-dir $env:binaryDir\src\java --build-config Release --verbose --timeout 10800

      - name: Run the Python Tests
        run: |
          python test/python/test_onnxruntime_genai.py --cwd "test\python" --test_models "test\test_models"

      - name: Build the C# API and Run the C# Tests
        run: |
          cd test\csharp
          dotnet test /p:NativeBuildOutputDir="$env:GITHUB_WORKSPACE\$env:binaryDir\Release" /p:OrtLibDir="$env:GITHUB_WORKSPACE\ort\lib"

      - name: Build the C# Examples
        run: |
          cd examples\csharp\ModelChat
          dotnet build -c Release
          cd ..\ModelMM
          dotnet build -c Release

      - name: Test the C# LLM Example with Tool Calling
        run: |
          python3 test\python\special_tokens.py -p test\test_models\qwen-2.5-0.5b\int4\cpu\tokenizer.json -s "<tool_call>" -e "</tool_call>"
          .\examples\csharp\ModelChat\bin\Release\net8.0\ModelChat.exe -m test\test_models\qwen-2.5-0.5b\int4\cpu\ -e cpu --response_format lark_grammar --tools_file test\test_models\tool-definitions\weather.json --tool_call_start "<tool_call>" --tool_call_end "</tool_call>" --user_prompt "What is the weather in Redmond, WA?" --tool_output --non_interactive --verbose

      - name: Verify Build Artifacts
        if: always()
        continue-on-error: true
        run: |
          Get-ChildItem -Path $env:GITHUB_WORKSPACE\$env:binaryDir -Recurse
          Get-ChildItem -Path $env:GITHUB_WORKSPACE\$env:binaryDir\test -Recurse

      - name: Run tests
        run: |-
          copy $env:GITHUB_WORKSPACE\ort\lib\* .\$env:binaryDir\Release
          & .\$env:binaryDir\Release\unit_tests.exe