Enable Wheel and NuGet packaging on macOS x64 & arm64

.github/workflows/linux-cpu-x64-build.yml

@@ -82,25 +82,28 @@ jobs:
         run: |
           rm -rf ort
 
+      - name: Verify Build Artifacts
+        if: always()
+        continue-on-error: true
+        run: |
+          ls -l ${{ github.workspace }}/build/cpu
+
       # This will also download all the test models to the test/test_models directory
       # These models are used by the python tests as well as C#, C++ and others.
       - name: Run the python tests
         run: |
+          export ORTGENAI_LOG_ORT_LIB=1
           python3 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: |
+          export ORTGENAI_LOG_ORT_LIB=1
           cd test/csharp
           dotnet test /p:Configuration=Release /p:NativeBuildOutputDir="../../build/cpu/"
 
-      - name: Verify Build Artifacts
-        if: always()
-        continue-on-error: true
-        run: |
-          ls -l ${{ github.workspace }}/build/cpu
-
       - name: Run tests
         run: |
           set -e -x
+          export ORTGENAI_LOG_ORT_LIB=1
           export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$GITHUB_WORKSPACE/build/cpu/
           ./build/cpu/test/unit_tests

.github/workflows/linux-gpu-x64-build.yml

@@ -154,4 +154,4 @@ jobs:
             --rm \
             --volume /data/ortgenai_pytorch_models:/data/ortgenai_pytorch_models \
             --volume $GITHUB_WORKSPACE:/ort_genai_src \
-            -w /ort_genai_src onnxruntimecudabuildx64 bash -c "LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/ort_genai_src/build/cuda/ /ort_genai_src/build/cuda/test/unit_tests"
+            -w /ort_genai_src onnxruntimecudabuildx64 bash -c "ORTGENAI_LOG_ORT_LIB=1 LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/ort_genai_src/build/cuda/ /ort_genai_src/build/cuda/test/unit_tests"

.github/workflows/mac-cpu-arm64-build.yml

@@ -14,7 +14,7 @@ env:
   ORT_PACKAGE_NAME: "Microsoft.ML.OnnxRuntime"
 jobs:
   mac-cpu-arm64-build:
-    runs-on: macos-latest
+    runs-on: macos-latest # arm64
     steps:
       - name: Checkout OnnxRuntime GenAI repo
         uses: actions/checkout@v4
@@ -36,26 +36,44 @@ jobs:
           mv ${{ env.ORT_PACKAGE_NAME }}/build/native/include ort/
           mv ${{ env.ORT_PACKAGE_NAME }}/runtimes/osx-arm64/native/* ort/lib/
 
-
       - name: Configure CMake
         run: |
-          cmake --preset macos_cpu_release
+          cmake --preset macos_arm64_cpu_release
 
       - name: Build with CMake
         run: |
-          cmake --build --preset macos_cpu_release --parallel
+          cmake --build --preset macos_arm64_cpu_release --parallel
         continue-on-error: false
 
+      - name: Install the python wheel and test dependencies
+        run: |
+          python3 -m venv genai-macos-venv
+          source genai-macos-venv/bin/activate
+          python3 -m pip install -r test/python/requirements.txt
+          python3 -m pip install -r test/python/requirements-macos.txt
+          python3 -m pip install build/cpu/osx-arm64/wheel/onnxruntime_genai*.whl
+
+      - name: Remove the ort lib and header files
+        run: |
+          rm -rf ort
+
       - name: Verify Build Artifacts
         if: always()
         continue-on-error: true
         run: |
-          ls -l ${{ github.workspace }}/build
-
-      - name: Upload Build Artifacts
-        uses: actions/upload-artifact@v3
-        with:
-          name: onnxruntime-genai-mac-cpu-arm64
-          path: ${{ github.workspace }}/build/**/*.a
+          ls -l ${{ github.workspace }}/build/cpu/osx-arm64
 
+      # This will also download all the test models to the test/test_models directory
+      # These models are used by the python tests as well as C#, C++ and others.
+      - name: Run the python tests
+        run: |
+          source genai-macos-venv/bin/activate
+          export ORTGENAI_LOG_ORT_LIB=1
+          python3 test/python/test_onnxruntime_genai.py --cwd test/python --test_models test/test_models
 
+      - name: Run tests
+        run: |
+          set -e -x
+          export ORTGENAI_LOG_ORT_LIB=1
+          export DYLD_LIBRARY_PATH=$DYLD_LIBRARY_PATH:$GITHUB_WORKSPACE/build/cpu/osx-arm64
+          ./build/cpu/osx-arm64/test/unit_tests

.pipelines/nuget-publishing.yml

@@ -33,6 +33,11 @@ parameters:
   type: boolean
   default: true
 
+- name: enable_macos_cpu
+  displayName: 'Whether MacOS CPU package is built.'
+  type: boolean
+  default: true
+
 - name: ort_version
   displayName: 'OnnxRuntime version'
   type: string
@@ -83,6 +88,7 @@ stages:
     enable_linux_cuda: ${{ parameters.enable_linux_cuda }}
     enable_win_dml: ${{ parameters.enable_win_dml }}
     enable_win_arm64: ${{ parameters.enable_win_arm64 }}
+    enable_macos_cpu: ${{ parameters.enable_macos_cpu }}
     ort_version: ${{ parameters.ort_version }}
     ort_cuda_version: ${{ parameters.ort_cuda_version }}
     ort_dml_version: ${{ parameters.ort_dml_version }}
@@ -97,6 +103,7 @@ stages:
     enable_linux_cuda: ${{ parameters.enable_linux_cuda }}
     enable_win_dml: ${{ parameters.enable_win_dml }}
     enable_win_arm64: ${{ parameters.enable_win_arm64 }}
+    enable_macos_cpu: ${{ parameters.enable_macos_cpu }}
     ort_version: ${{ parameters.ort_version }}
     ort_cuda_version: ${{ parameters.ort_cuda_version }}
     ort_dml_version: ${{ parameters.ort_dml_version }}
@@ -111,6 +118,7 @@ stages:
       enable_linux_cuda: ${{ parameters.enable_linux_cuda }}
       enable_win_dml: ${{ parameters.enable_win_dml }}
       enable_win_arm64: ${{ parameters.enable_win_arm64 }}
+      enable_macos_cpu: ${{ parameters.enable_macos_cpu }}
       ort_version: ${{ parameters.ort_version }}
       ort_cuda_version: ${{ parameters.ort_cuda_version }}
       ort_dml_version: ${{ parameters.ort_dml_version }}

.pipelines/pypl-publishing.yml

@@ -39,6 +39,11 @@ parameters:
   type: boolean
   default: true
 
+- name: enable_macos_cpu
+  displayName: 'Whether MacOS CPU package is built.'
+  type: boolean
+  default: true
+
 - name: ort_version
   displayName: 'OnnxRuntime version'
   type: string
@@ -97,6 +102,7 @@ stages:
     enable_win_cuda: ${{ parameters.enable_win_cuda }}
     enable_win_dml: ${{ parameters.enable_win_dml }}
     enable_win_arm64_cpu: ${{ parameters.enable_win_arm64_cpu }}
+    enable_macos_cpu: ${{ parameters.enable_macos_cpu }}
     ort_version: ${{ parameters.ort_version }}
     ort_cuda_118_version: ${{ parameters.ort_cuda_118_version }}
     ort_cuda_122_version: ${{ parameters.ort_cuda_122_version }}
@@ -113,6 +119,7 @@ stages:
       enable_win_cuda: ${{ parameters.enable_win_cuda }}
       enable_win_dml: ${{ parameters.enable_win_dml }}
       enable_win_arm64_cpu: ${{ parameters.enable_win_arm64_cpu }}
+      enable_macos_cpu: ${{ parameters.enable_macos_cpu }}
       ort_version: ${{ parameters.ort_version }}
       ort_cuda_118_version: ${{ parameters.ort_cuda_118_version }}
       ort_cuda_122_version: ${{ parameters.ort_cuda_122_version }}

.pipelines/stages/capi-packaging-stage.yml

@@ -11,6 +11,8 @@ parameters:
   type: boolean
 - name: enable_linux_cuda
   type: boolean
+- name: enable_macos_cpu
+  type: boolean
 - name: ort_version
   type: string
 - name: ort_cuda_version
@@ -91,4 +93,21 @@ stages:
         ep: 'cuda'
         ort_version: ${{ parameters.ort_cuda_version }}
         os: 'linux'
-        build_config: ${{ parameters.build_config }}
+        build_config: ${{ parameters.build_config }}
+
+  - ${{ if eq(parameters.enable_macos_cpu, true) }}:
+    - template: jobs/capi-packaging-job.yml
+      parameters:
+        arch: 'x64'
+        ep: 'cpu'
+        ort_version: ${{ parameters.ort_version }}
+        os: 'osx'
+        build_config: ${{ parameters.build_config }}
+
+    - template: jobs/capi-packaging-job.yml
+      parameters:
+        arch: 'arm64'
+        ep: 'cpu'
+        ort_version: ${{ parameters.ort_version }}
+        os: 'osx'
+        build_config: ${{ parameters.build_config }}

.pipelines/stages/jobs/capi-packaging-job.yml

@@ -13,6 +13,7 @@ parameters:
   values:
   - 'linux'
   - 'win'
+  - 'osx'
 - name: build_config
   type: string
   default: 'release'
@@ -33,6 +34,10 @@ jobs:
       pool: 'onnxruntime-genai-windows-vs-2022-arm64'
     ${{ else }}:
       pool: 'onnxruntime-Win-CPU-2022'
+  ${{ if eq(parameters.os, 'osx') }}:
+    pool:
+      vmImage: 'macOS-latest'
+
   timeoutInMinutes: 180
   #  set variables here to be used in the template and steps
   variables:
@@ -130,4 +135,12 @@ jobs:
         ep: ${{ parameters.ep }}
         build_config: ${{ parameters.build_config }}
 
+  - ${{ if eq(parameters.os, 'osx') }}:
+    - template: steps/capi-macos-step.yml
+      parameters:
+        target: 'onnxruntime-genai'
+        arch: ${{ parameters.arch }}
+        ep: ${{ parameters.ep }}
+        build_config: ${{ parameters.build_config }}
+
   - template: steps/compliant-and-cleanup-step.yml

.pipelines/stages/jobs/nuget-packaging-job.yml

@@ -32,6 +32,11 @@ parameters:
   type: boolean
   default: false
 
+- name: enable_macos_cpu
+  displayName: 'Whether MacOS CPU package is built.'
+  type: boolean
+  default: false
+
 - name: ort_version
   type: string
 
@@ -130,6 +135,18 @@ jobs:
         ArtifactName: onnxruntime-genai-linux-cuda-x64-capi
         TargetPath: '$(Build.BinariesDirectory)/artifact-downloads'
 
+  - ${{ if eq(parameters.enable_macos_cpu, true) }}:
+    - template: steps/utils/flex-download-pipeline-artifact.yml
+      parameters:
+        StepName: 'Download osx-cpu-x64 capi Artifacts'
+        ArtifactName: onnxruntime-genai-osx-cpu-x64-capi
+        TargetPath: '$(Build.BinariesDirectory)/artifact-downloads'
+    - template: steps/utils/flex-download-pipeline-artifact.yml
+      parameters:
+        StepName: 'Download osx-cpu-arm64 capi Artifacts'
+        ArtifactName: onnxruntime-genai-osx-cpu-arm64-capi
+        TargetPath: '$(Build.BinariesDirectory)/artifact-downloads'
+
   - checkout: self
     path: onnxruntime-genai
     clean: true

.pipelines/stages/jobs/nuget-validation-job.yml

@@ -21,6 +21,7 @@ parameters:
   values:
   - 'linux'
   - 'win'
+  - 'osx'
 
 jobs:
 - job: nuget_${{ parameters.os }}_${{ parameters.ep }}_${{ parameters.arch }}_validation
@@ -43,6 +44,9 @@ jobs:
       pool: 'onnxruntime-genai-windows-vs-2022-arm64'
     ${{ else }}:
       pool: 'onnxruntime-Win-CPU-2022'
+  ${{ if eq(parameters.os, 'osx') }}:
+    pool:
+      vmImage: 'macOS-latest'
 
   timeoutInMinutes: 180
   #  set variables here to be used in the template and steps
@@ -158,14 +162,15 @@ jobs:
       env:
         NUGET_PLUGIN_HANDSHAKE_TIMEOUT_IN_SECONDS: 180
         NUGET_PLUGIN_REQUEST_TIMEOUT_IN_SECONDS: 180
-  - ${{ elseif eq(parameters.os, 'linux') }}:
+  - ${{ elseif or(eq(parameters.os, 'linux'), eq(parameters.os, 'osx')) }}:
     - bash: |
        dotnet --info
        cp $(Build.BinariesDirectory)/nuget/* examples/csharp/HelloPhi/
        cd examples/csharp/HelloPhi
        mv models/$(prebuild_phi3_mini_model_folder) models/phi-3
        dotnet restore -r $(os)-$(arch) /property:Configuration=$(csproj_configuration) --source https://api.nuget.org/v3/index.json --source https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/ORT-Nightly/nuget/v3/index.json --source $PWD --disable-parallel --verbosity detailed
        dotnet build ./HelloPhi.csproj -r $(os)-$(arch) /property:Configuration=$(csproj_configuration) --no-restore --self-contained
+       ls -l ./bin/$(csproj_configuration)/net6.0/$(os)-$(arch)/
       displayName: 'Perform dotnet restore & build'
       workingDirectory: '$(Build.Repository.LocalPath)'
       env:

.pipelines/stages/jobs/py-packaging-job.yml

@@ -16,6 +16,7 @@ parameters:
   values:
   - 'linux'
   - 'win'
+  - 'osx'
 - name: build_config
   type: string
   default: 'release'
@@ -37,9 +38,12 @@ jobs:
       pool: 'onnxruntime-genai-windows-vs-2022-arm64'
     ${{ else }}:
       pool: 'onnxruntime-Win-CPU-2022'
+  ${{ if eq(parameters.os, 'osx') }}:
+    pool:
+      vmImage: 'macOS-latest'
 
   strategy:
-    ${{ if eq(parameters.arch, 'arm64') }}:
+    ${{ if and(eq(parameters.os, 'win'), eq(parameters.arch, 'arm64')) }}: # Older Python versions don't have official win-arm64 build.
       matrix:
         Python311:
           PyDotVer: '3.11'
@@ -73,6 +77,8 @@ jobs:
   - name: skipComponentGovernanceDetection
     ${{ if eq(parameters.os, 'linux') }}:
       value: true
+    ${{ if eq(parameters.os, 'osx') }}:
+      value: true
     ${{ if eq(parameters.os, 'win') }}:
       value: false
   - name: arch
@@ -179,5 +185,13 @@ jobs:
         ep: ${{ parameters.ep }}
         build_config: ${{ parameters.build_config }}
 
+  - ${{ if eq(parameters.os, 'osx') }}:
+    - template: steps/capi-macos-step.yml
+      parameters:
+        target: 'python'
+        arch: ${{ parameters.arch }}
+        ep: ${{ parameters.ep }}
+        build_config: ${{ parameters.build_config }}
+
   - template: steps/compliant-and-cleanup-step.yml