From 39a0da19cb6a042218a604609b3f680dfc53440a Mon Sep 17 00:00:00 2001 From: Chester Liu <4710575+skyline75489@users.noreply.github.com> Date: Thu, 12 Sep 2024 11:02:40 +0800 Subject: [PATCH] Enable Wheel and NuGet packaging on macOS x64 & arm64 (#841) * Implement #693 on macOS for non-framework usage. * Enable macOS x64 & ARM64 wheel build & packaging * Enable macOS x64 & ARM64 NuGet build & packaging Python validation run: * https://aiinfra.visualstudio.com/ONNX%20Runtime/_build/results?buildId=548627&view=results NuGet validation run: * https://aiinfra.visualstudio.com/ONNX%20Runtime/_build/results?buildId=548631&view=results --- .github/workflows/linux-cpu-x64-build.yml | 15 +-- .github/workflows/linux-gpu-x64-build.yml | 2 +- .github/workflows/mac-cpu-arm64-build.yml | 44 +++++++-- .pipelines/nuget-publishing.yml | 8 ++ .pipelines/pypl-publishing.yml | 7 ++ .pipelines/stages/capi-packaging-stage.yml | 21 +++- .pipelines/stages/jobs/capi-packaging-job.yml | 13 +++ .../stages/jobs/nuget-packaging-job.yml | 17 ++++ .../stages/jobs/nuget-validation-job.yml | 7 +- .pipelines/stages/jobs/py-packaging-job.yml | 16 ++- .pipelines/stages/jobs/py-validation-job.yml | 36 +++---- .../stages/jobs/steps/capi-linux-step.yml | 2 +- .../stages/jobs/steps/capi-macos-step.yml | 99 +++++++++++++++++++ .../jobs/steps/compliant-and-cleanup-step.yml | 1 + .../utils/download-huggingface-model.yml | 2 +- .pipelines/stages/nuget-packaging-stage.yml | 8 +- .pipelines/stages/nuget-validation-stage.yml | 12 +++ .pipelines/stages/py-packaging-stage.yml | 19 +++- .pipelines/stages/py-validation-stage.yml | 12 +++ CMakeLists.txt | 6 +- cmake/package.cmake | 6 +- cmake/presets/CMakeMacOSBuildPresets.json | 8 +- cmake/presets/CMakeMacOSConfigPresets.json | 25 ++++- examples/csharp/HelloPhi/Program.cs | 1 + src/models/onnxruntime_api.h | 59 +++++++---- src/python/CMakeLists.txt | 2 +- src/python/py/_dll_directory.py | 16 ++- test/CMakeLists.txt | 2 +- ...Microsoft.ML.OnnxRuntimeGenAI.Tests.csproj | 10 ++ test/python/requirements-macos.txt | 4 + .../nuget/generate_nuspec_for_native_nuget.py | 4 +- 31 files changed, 410 insertions(+), 74 deletions(-) create mode 100644 .pipelines/stages/jobs/steps/capi-macos-step.yml create mode 100644 test/python/requirements-macos.txt diff --git a/.github/workflows/linux-cpu-x64-build.yml b/.github/workflows/linux-cpu-x64-build.yml index 6f970fdc0..5594cc5e1 100644 --- a/.github/workflows/linux-cpu-x64-build.yml +++ b/.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 diff --git a/.github/workflows/linux-gpu-x64-build.yml b/.github/workflows/linux-gpu-x64-build.yml index 857d54fd1..a675b1037 100644 --- a/.github/workflows/linux-gpu-x64-build.yml +++ b/.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" diff --git a/.github/workflows/mac-cpu-arm64-build.yml b/.github/workflows/mac-cpu-arm64-build.yml index 9f861a96d..a5f969284 100644 --- a/.github/workflows/mac-cpu-arm64-build.yml +++ b/.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 +36,50 @@ 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 --no-deps + + - 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 + ls -l ${{ github.workspace }}/build/cpu/osx-arm64 - - name: Upload Build Artifacts - uses: actions/upload-artifact@v3 - with: - name: onnxruntime-genai-mac-cpu-arm64 - path: ${{ github.workspace }}/build/**/*.a + # This will also download all the test models to the test/test_models directory + # These models are used by the python tests as well as C#, C++ and others. + - name: Run the python tests + run: | + source genai-macos-venv/bin/activate + export ORTGENAI_LOG_ORT_LIB=1 + python3 test/python/test_onnxruntime_genai.py --cwd test/python --test_models test/test_models + - name: Build the C# API and Run the C# Tests + run: | + export ORTGENAI_LOG_ORT_LIB=1 + cd test/csharp + dotnet test /p:Configuration=Release /p:NativeBuildOutputDir="../../build/cpu/osx-arm64" + - name: Run tests + run: | + set -e -x + export ORTGENAI_LOG_ORT_LIB=1 + export DYLD_LIBRARY_PATH=$DYLD_LIBRARY_PATH:$GITHUB_WORKSPACE/build/cpu/osx-arm64 + ./build/cpu/osx-arm64/test/unit_tests diff --git a/.pipelines/nuget-publishing.yml b/.pipelines/nuget-publishing.yml index 137e2c36f..726493826 100644 --- a/.pipelines/nuget-publishing.yml +++ b/.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 }} diff --git a/.pipelines/pypl-publishing.yml b/.pipelines/pypl-publishing.yml index 6ecbc938e..099679cc3 100644 --- a/.pipelines/pypl-publishing.yml +++ b/.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 }} diff --git a/.pipelines/stages/capi-packaging-stage.yml b/.pipelines/stages/capi-packaging-stage.yml index 1b9021a72..2da3971a8 100644 --- a/.pipelines/stages/capi-packaging-stage.yml +++ b/.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 }} \ No newline at end of file + 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 }} diff --git a/.pipelines/stages/jobs/capi-packaging-job.yml b/.pipelines/stages/jobs/capi-packaging-job.yml index 1d936eccb..aa9e9efa4 100644 --- a/.pipelines/stages/jobs/capi-packaging-job.yml +++ b/.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: ${{ parameters.build_config }} + - template: steps/compliant-and-cleanup-step.yml diff --git a/.pipelines/stages/jobs/nuget-packaging-job.yml b/.pipelines/stages/jobs/nuget-packaging-job.yml index e01374b9e..8599ee207 100644 --- a/.pipelines/stages/jobs/nuget-packaging-job.yml +++ b/.pipelines/stages/jobs/nuget-packaging-job.yml @@ -32,6 +32,11 @@ parameters: type: boolean default: false +- name: enable_macos_cpu + displayName: 'Whether MacOS CPU package is built.' + type: boolean + default: false + - name: ort_version type: string @@ -130,6 +135,18 @@ jobs: ArtifactName: onnxruntime-genai-linux-cuda-x64-capi TargetPath: '$(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: '$(Build.BinariesDirectory)/artifact-downloads' + - checkout: self path: onnxruntime-genai clean: true diff --git a/.pipelines/stages/jobs/nuget-validation-job.yml b/.pipelines/stages/jobs/nuget-validation-job.yml index e382a3ca7..c5498e12e 100644 --- a/.pipelines/stages/jobs/nuget-validation-job.yml +++ b/.pipelines/stages/jobs/nuget-validation-job.yml @@ -21,6 +21,7 @@ parameters: values: - 'linux' - 'win' + - 'osx' jobs: - job: nuget_${{ 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 @@ -158,7 +162,7 @@ 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 $(Build.BinariesDirectory)/nuget/* examples/csharp/HelloPhi/ @@ -166,6 +170,7 @@ jobs: mv models/$(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: diff --git a/.pipelines/stages/jobs/py-packaging-job.yml b/.pipelines/stages/jobs/py-packaging-job.yml index 99582c9f1..db39b9afb 100644 --- a/.pipelines/stages/jobs/py-packaging-job.yml +++ b/.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: ${{ parameters.build_config }} + - template: steps/compliant-and-cleanup-step.yml diff --git a/.pipelines/stages/jobs/py-validation-job.yml b/.pipelines/stages/jobs/py-validation-job.yml index 6e3bd6625..2282f8f77 100644 --- a/.pipelines/stages/jobs/py-validation-job.yml +++ b/.pipelines/stages/jobs/py-validation-job.yml @@ -24,6 +24,7 @@ parameters: values: - 'linux' - 'win' + - 'osx' jobs: - job: python_${{ parameters.os }}_${{ parameters.ep }}${{ parameters.cuda_display_version }}_${{ parameters.arch }}_validation @@ -46,6 +47,9 @@ jobs: pool: 'onnxruntime-genai-windows-vs-2022-arm64' ${{ else }}: pool: 'onnxruntime-Win-CPU-2022' + ${{ if eq(parameters.os, 'osx') }}: + pool: + vmImage: 'macOS-latest' timeoutInMinutes: 240 workspace: @@ -147,22 +151,13 @@ jobs: addToPath: true architecture: $(arch) - - ${{ if eq(parameters.os, 'linux') }}: - - template: steps/utils/flex-download-pipeline-artifact.yml - parameters: - StepName: 'Download Python Wheel Artifacts' - ArtifactName: $(ArtifactName) - TargetPath: '$(Build.BinariesDirectory)/wheel' - SpecificArtifact: ${{ parameters.specificArtifact }} - BuildId: ${{ parameters.BuildId }} - - ${{ if eq(parameters.os, 'win') }}: - - template: steps/utils/flex-download-pipeline-artifact.yml - parameters: - StepName: 'Download Python Wheel Artifacts' - ArtifactName: $(ArtifactName)-wheel - TargetPath: '$(Build.BinariesDirectory)/wheel' - SpecificArtifact: ${{ parameters.specificArtifact }} - BuildId: ${{ parameters.BuildId }} + - template: steps/utils/flex-download-pipeline-artifact.yml + parameters: + StepName: 'Download Python Wheel Artifacts' + ArtifactName: $(ArtifactName)-wheel + TargetPath: '$(Build.BinariesDirectory)/wheel' + SpecificArtifact: ${{ parameters.specificArtifact }} + BuildId: ${{ parameters.BuildId }} - template: steps/utils/download-huggingface-model.yml parameters: @@ -174,7 +169,7 @@ jobs: HuggingFaceToken: $(HF_TOKEN) os: ${{ parameters.os }} - - ${{ if eq(parameters.os, 'linux') }}: + - ${{ if or(eq(parameters.os, 'linux'), eq(parameters.os, 'osx')) }}: - ${{ if eq(parameters.ep, 'cuda') }}: - bash: | set -e -x @@ -204,7 +199,12 @@ jobs: - bash: | export ORTGENAI_LOG_ORT_LIB=1 python -m pip install -r test/python/requirements.txt - python -m pip install -r test/python/requirements-cpu.txt + if [[ "$(os)" == "linux" ]]; then + python -m pip install -r test/python/requirements-cpu.txt + fi + if [[ "$(os)" == "osx" ]]; then + python -m pip install -r test/python/requirements-macos.txt + fi cd examples/python python -m pip install --no-index --find-links=$(Build.BinariesDirectory)/wheel $(pip_package_name) python model-generate.py -m ./models/$(prebuild_phi3_mini_model_folder) --min_length 25 --max_length 50 --verbose diff --git a/.pipelines/stages/jobs/steps/capi-linux-step.yml b/.pipelines/stages/jobs/steps/capi-linux-step.yml index c3074f607..de7e5ae6c 100644 --- a/.pipelines/stages/jobs/steps/capi-linux-step.yml +++ b/.pipelines/stages/jobs/steps/capi-linux-step.yml @@ -169,7 +169,7 @@ steps: - task: PublishBuildArtifacts@1 displayName: 'Publish Artifact: ONNXRuntime python wheel' inputs: - ArtifactName: $(artifactName) + ArtifactName: $(artifactName)-wheel PathtoPublish: '$(Build.ArtifactStagingDirectory)/wheel' - script: | diff --git a/.pipelines/stages/jobs/steps/capi-macos-step.yml b/.pipelines/stages/jobs/steps/capi-macos-step.yml new file mode 100644 index 000000000..73818fdaf --- /dev/null +++ b/.pipelines/stages/jobs/steps/capi-macos-step.yml @@ -0,0 +1,99 @@ +parameters: +- name: target + type: string +- name: ep + type: string + default: 'cpu' +- name: arch + type: string + default: 'x64' +- name: build_config + type: string + default: 'release' + +steps: + +- checkout: self + clean: true + path: onnxruntime-genai + submodules: recursive + +- template: utils/set-nightly-build-option-variable.yml + +- bash: | + echo "arch=$(arch)" + echo "ort_filename=$(ort_filename)" + echo "ort_version=$(ort_version)" + echo "ep=$(ep)" + echo "cuda_version=$(cuda_version)" + echo "target=${{ parameters.target }}" + echo "build_config=${{ parameters.build_config }}" + displayName: 'Print Parameters' + +- template: utils/download-ort.yml + parameters: + archiveType: 'zip' + +- powershell: | + $env:MACOSX_DEPLOYMENT_TARGET = "12.0" # Monterey + cmake --preset macos_$(arch)_$(ep)_$(build_config) + displayName: 'Configure CMake C API' + workingDirectory: '$(Build.Repository.LocalPath)' + +- powershell: | + cmake --build --preset macos_$(arch)_$(ep)_$(build_config) --parallel --target ${{ parameters.target }} + displayName: 'Build C API' + workingDirectory: '$(Build.Repository.LocalPath)' + +- ${{ if eq(parameters.target, 'onnxruntime-genai') }}: + - powershell: | + python -m pip install wheel + displayName: 'Install wheel' + + - powershell: | + cmake --build --preset macos_$(arch)_$(ep)_$(build_config) --target package + displayName: 'Package C/C++ API' + workingDirectory: '$(Build.Repository.LocalPath)' + + - task: PublishBuildArtifacts@1 + displayName: 'Publish Artifact: ONNXRuntime Genai capi' + inputs: + ArtifactName: $(artifactName)-capi + PathtoPublish: '$(Build.Repository.LocalPath)/build/$(ep)/$(os)-$(arch)/package' + +- ${{ if eq(parameters.target, 'python') }}: + - powershell: | + python -m pip install wheel + displayName: 'Install wheel' + + - powershell: | + # From: https://github.com/pypa/cibuildwheel/blob/93542c397cfe940bcbb8f1eff5c37d345ea16653/cibuildwheel/macos.py#L247-L260 + if ("$(arch)" -eq "arm64") { + $env:_PYTHON_HOST_PLATFORM = "macosx-12.0-arm64" + $env:ARCHFLAGS = "-arch arm64" + } + else { + $env:_PYTHON_HOST_PLATFORM = "macosx-12.0-x86_64" + $env:ARCHFLAGS = "-arch x86_64" + } + cmake --build --preset macos_$(arch)_$(ep)_$(build_config) --parallel --PyPackageBuild + displayName: 'Build Python Wheel' + workingDirectory: '$(Build.Repository.LocalPath)' + + - powershell: | + Get-ChildItem -Path $(Build.Repository.LocalPath) -Recurse + displayName: 'List all files in the repo for debugging' + workingDirectory: '$(Build.Repository.LocalPath)' + + - task: CopyFiles@2 + displayName: 'Copy Python Wheel to: $(Build.ArtifactStagingDirectory)' + inputs: + SourceFolder: '$(Build.Repository.LocalPath)/build/$(ep)/$(os)-$(arch)/wheel' + Contents: '*.whl' + TargetFolder: '$(Build.ArtifactStagingDirectory)/wheel' + + - task: PublishBuildArtifacts@1 + displayName: 'Publish Artifact: ONNXRuntime python wheel' + inputs: + ArtifactName: $(artifactName)-wheel + PathtoPublish: '$(Build.ArtifactStagingDirectory)/wheel' diff --git a/.pipelines/stages/jobs/steps/compliant-and-cleanup-step.yml b/.pipelines/stages/jobs/steps/compliant-and-cleanup-step.yml index aa2f7b547..19ee48dcc 100644 --- a/.pipelines/stages/jobs/steps/compliant-and-cleanup-step.yml +++ b/.pipelines/stages/jobs/steps/compliant-and-cleanup-step.yml @@ -4,6 +4,7 @@ steps: inputs: debugMode: false continueOnError: true + condition: ne(variables['os'], 'osx') # Not available on macOS. See https://eng.ms/docs/products/credential-risk-exposure-defense/solutions/credscan_client/overview - task: TSAUpload@2 displayName: 'TSA upload' diff --git a/.pipelines/stages/jobs/steps/utils/download-huggingface-model.yml b/.pipelines/stages/jobs/steps/utils/download-huggingface-model.yml index 30014e6ff..0dc5a8fc1 100644 --- a/.pipelines/stages/jobs/steps/utils/download-huggingface-model.yml +++ b/.pipelines/stages/jobs/steps/utils/download-huggingface-model.yml @@ -15,7 +15,7 @@ parameters: type: string steps: - - ${{ if eq(parameters.os, 'linux') }}: + - ${{ if or(eq(parameters.os, 'linux'), eq(parameters.os, 'osx')) }}: - bash: | python -m pip install "huggingface_hub[cli]" huggingface-cli login --token $HF_TOKEN diff --git a/.pipelines/stages/nuget-packaging-stage.yml b/.pipelines/stages/nuget-packaging-stage.yml index 4137b29bd..00ad62dd2 100644 --- a/.pipelines/stages/nuget-packaging-stage.yml +++ b/.pipelines/stages/nuget-packaging-stage.yml @@ -29,6 +29,11 @@ parameters: type: boolean default: true +- name: enable_macos_cpu + displayName: 'Whether MacOS CPU package is built.' + type: boolean + default: true + - name: ort_version type: string - name: ort_cuda_version @@ -42,7 +47,7 @@ parameters: stages: - stage: nuget_packaging jobs: - - ${{ if or(eq(parameters.enable_linux_cpu, true), eq(parameters.enable_win_cpu, true)) }}: + - ${{ if or(eq(parameters.enable_linux_cpu, true), eq(parameters.enable_win_cpu, true), eq(parameters.enable_macos_cpu, true)) }}: - template: jobs/nuget-packaging-job.yml parameters: ep: 'cpu' @@ -51,6 +56,7 @@ stages: enable_linux_cpu: ${{ parameters.enable_linux_cpu }} enable_win_cpu: ${{ parameters.enable_win_cpu }} enable_win_arm64: ${{ parameters.enable_win_arm64 }} + enable_macos_cpu: ${{ parameters.enable_macos_cpu }} - ${{ if or(eq(parameters.enable_linux_cuda, true), eq(parameters.enable_win_cuda, true)) }}: - template: jobs/nuget-packaging-job.yml parameters: diff --git a/.pipelines/stages/nuget-validation-stage.yml b/.pipelines/stages/nuget-validation-stage.yml index 380fda54b..d8c8b3ca4 100644 --- a/.pipelines/stages/nuget-validation-stage.yml +++ b/.pipelines/stages/nuget-validation-stage.yml @@ -19,6 +19,8 @@ parameters: type: boolean - name: enable_linux_cuda type: boolean +- name: enable_macos_cpu + type: boolean - name: ort_version type: string - name: ort_cuda_version @@ -104,3 +106,13 @@ stages: os: 'linux' SpecificArtifact: ${{ parameters.SpecificArtifact }} BuildId: ${{ parameters.BuildId }} + + - ${{ if eq(parameters.enable_macos_cpu, true) }}: + - template: jobs/nuget-validation-job.yml + parameters: + arch: 'x64' + ep: 'cpu' + ort_version: ${{ parameters.ort_version }} + os: 'osx' + SpecificArtifact: ${{ parameters.SpecificArtifact }} + BuildId: ${{ parameters.BuildId }} \ No newline at end of file diff --git a/.pipelines/stages/py-packaging-stage.yml b/.pipelines/stages/py-packaging-stage.yml index d43868ed0..d7bab62c8 100644 --- a/.pipelines/stages/py-packaging-stage.yml +++ b/.pipelines/stages/py-packaging-stage.yml @@ -13,6 +13,8 @@ parameters: type: boolean - name: enable_linux_rocm type: boolean +- name: enable_macos_cpu + type: boolean - name: ort_version type: string - name: ort_cuda_118_version @@ -121,5 +123,18 @@ stages: ort_version: ${{ parameters.ort_rocm_version }} os: 'linux' build_config: ${{ parameters.build_config }} - - + - ${{ if eq(parameters.enable_macos_cpu, true) }}: + - template: jobs/py-packaging-job.yml + parameters: + arch: 'x64' + ep: 'cpu' + ort_version: ${{ parameters.ort_version }} + os: 'osx' + build_config: ${{ parameters.build_config }} + - template: jobs/py-packaging-job.yml + parameters: + arch: 'arm64' + ep: 'cpu' + ort_version: ${{ parameters.ort_version }} + os: 'osx' + build_config: ${{ parameters.build_config }} diff --git a/.pipelines/stages/py-validation-stage.yml b/.pipelines/stages/py-validation-stage.yml index d619c3acb..0b6abda5a 100644 --- a/.pipelines/stages/py-validation-stage.yml +++ b/.pipelines/stages/py-validation-stage.yml @@ -19,6 +19,8 @@ parameters: type: boolean - name: enable_linux_cuda type: boolean +- name: enable_macos_cpu + type: boolean - name: ort_version type: string - name: ort_cuda_118_version @@ -123,3 +125,13 @@ stages: os: 'linux' SpecificArtifact: ${{ parameters.SpecificArtifact }} BuildId: ${{ parameters.BuildId }} + + - ${{ if eq(parameters.enable_macos_cpu, true) }}: + - template: jobs/py-validation-job.yml + parameters: + arch: 'x64' + ep: 'cpu' + ort_version: ${{ parameters.ort_version }} + os: 'osx' + SpecificArtifact: ${{ parameters.SpecificArtifact }} + BuildId: ${{ parameters.BuildId }} \ No newline at end of file diff --git a/CMakeLists.txt b/CMakeLists.txt index 06d196d7b..a35772f4c 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -84,8 +84,10 @@ target_link_directories(onnxruntime-genai PRIVATE ${ORT_LIB_DIR}) # we keep the shared libraries disconnected on Android as they will come from separate AARs and we don't want to force # the ORT version to match in both. -if(NOT (CMAKE_SYSTEM_NAME STREQUAL "Android" OR CMAKE_SYSTEM_NAME STREQUAL "Linux")) -target_link_libraries(onnxruntime-genai PRIVATE ${ONNXRUNTIME_LIB}) +if(CMAKE_SYSTEM_NAME STREQUAL "Android" OR CMAKE_SYSTEM_NAME STREQUAL "Linux" OR (CMAKE_SYSTEM_NAME STREQUAL "Darwin" AND (NOT BUILD_APPLE_FRAMEWORK))) + add_compile_definitions(_ORT_GENAI_USE_DLOPEN) +else() + target_link_libraries(onnxruntime-genai PRIVATE ${ONNXRUNTIME_LIB}) endif() set_target_properties(onnxruntime-genai PROPERTIES FOLDER "Sources") diff --git a/cmake/package.cmake b/cmake/package.cmake index 87ad79345..5939b1078 100644 --- a/cmake/package.cmake +++ b/cmake/package.cmake @@ -47,7 +47,11 @@ elseif (LINUX) set(CPACK_PACKAGE_FILE_NAME "onnxruntime-genai-${VERSION_INFO}-linux-arm64") endif () elseif (APPLE) - set(CPACK_PACKAGE_FILE_NAME "onnxruntime-genai-${VERSION_INFO}-osx-arm64") + if (CMAKE_OSX_ARCHITECTURES STREQUAL "x86_64") + set(CPACK_PACKAGE_FILE_NAME "onnxruntime-genai-${VERSION_INFO}-osx-x64") + else () + set(CPACK_PACKAGE_FILE_NAME "onnxruntime-genai-${VERSION_INFO}-osx-arm64") + endif() endif () if (WIN32) diff --git a/cmake/presets/CMakeMacOSBuildPresets.json b/cmake/presets/CMakeMacOSBuildPresets.json index e27635e9c..d3634c3d2 100644 --- a/cmake/presets/CMakeMacOSBuildPresets.json +++ b/cmake/presets/CMakeMacOSBuildPresets.json @@ -5,8 +5,12 @@ ], "buildPresets": [ { - "name": "macos_cpu_release", - "configurePreset": "macos_cpu_release" + "name": "macos_x64_cpu_release", + "configurePreset": "macos_x64_cpu_release" + }, + { + "name": "macos_arm64_cpu_release", + "configurePreset": "macos_arm64_cpu_release" }, { "name": "macos_cpu_debug", diff --git a/cmake/presets/CMakeMacOSConfigPresets.json b/cmake/presets/CMakeMacOSConfigPresets.json index a425dc595..7de4f39a9 100644 --- a/cmake/presets/CMakeMacOSConfigPresets.json +++ b/cmake/presets/CMakeMacOSConfigPresets.json @@ -4,13 +4,12 @@ "CMakeLinuxDefaultConfigPresets.json" ], "configurePresets": [ - { + { "name": "macos_default", "generator": "Unix Makefiles", "binaryDir": "${sourceDir}/build/cpu", "cacheVariables": { "CMAKE_POSITION_INDEPENDENT_CODE": "ON", - "CMAKE_OSX_ARCHITECTURES": "arm64", "USE_CUDA": "OFF", "USE_ROCM": "OFF" }, @@ -87,6 +86,28 @@ "macos_default", "macos_minsizerel_default" ] + }, + { + "name": "macos_x64_cpu_release", + "inherits": [ + "macos_cpu_release" + ], + "cacheVariables": { + "CMAKE_OSX_ARCHITECTURES": "x86_64" + }, + "displayName": "macos x64 cpu release", + "binaryDir": "${sourceDir}/build/cpu/osx-x64" + }, + { + "name": "macos_arm64_cpu_release", + "inherits": [ + "macos_cpu_release" + ], + "cacheVariables": { + "CMAKE_OSX_ARCHITECTURES": "arm64" + }, + "displayName": "macos arm64 cpu release", + "binaryDir": "${sourceDir}/build/cpu/osx-arm64" } ] } \ No newline at end of file diff --git a/examples/csharp/HelloPhi/Program.cs b/examples/csharp/HelloPhi/Program.cs index 82576c1ad..26e20a353 100644 --- a/examples/csharp/HelloPhi/Program.cs +++ b/examples/csharp/HelloPhi/Program.cs @@ -76,6 +76,7 @@ void PrintUsage() var sequences = tokenizer.Encode($"<|user|>{prompt}<|end|><|assistant|>"); using GeneratorParams generatorParams = new GeneratorParams(model); + generatorParams.SetSearchOption("min_length", 50); generatorParams.SetSearchOption("max_length", 200); generatorParams.SetInputSequences(sequences); if (option == 1) // Complete Output diff --git a/src/models/onnxruntime_api.h b/src/models/onnxruntime_api.h index ddc617f79..5466c7cb7 100644 --- a/src/models/onnxruntime_api.h +++ b/src/models/onnxruntime_api.h @@ -75,10 +75,11 @@ p_session_->Run(nullptr, input_names, inputs, std::size(inputs), output_names, o #include "../logging.h" #include "env_utils.h" -#if defined(__ANDROID__) -#include +#if defined(__linux__) #include +#if defined(__ANDROID__) +#include #define TAG "GenAI" #define LOG_DEBUG(...) __android_log_print(ANDROID_LOG_DEBUG, TAG, __VA_ARGS__) @@ -86,9 +87,15 @@ p_session_->Run(nullptr, input_names, inputs, std::size(inputs), output_names, o #define LOG_WARN(...) __android_log_print(ANDROID_LOG_WARN, TAG, __VA_ARGS__) #define LOG_ERROR(...) __android_log_print(ANDROID_LOG_ERROR, TAG, __VA_ARGS__) #define LOG_FATAL(...) __android_log_print(ANDROID_LOG_FATAL, TAG, __VA_ARGS__) +#endif -#elif defined(__linux__) +#elif defined(__APPLE__) +#include "TargetConditionals.h" +#if TARGET_OS_OSX && _ORT_GENAI_USE_DLOPEN +#define MACOS_USE_DLOPEN #include +#endif +#endif #ifndef PATH_MAX #define PATH_MAX (4096) @@ -103,8 +110,6 @@ p_session_->Run(nullptr, input_names, inputs, std::size(inputs), output_names, o #define LOG_ERROR(...) LOG_WHEN_ENABLED(Generators::Log("error", __VA_ARGS__)) #define LOG_FATAL(...) LOG_WHEN_ENABLED(Generators::Log("fatal", __VA_ARGS__)) -#endif - /** \brief Free functions and a few helpers are defined inside this namespace. Otherwise all types are the C API types * */ @@ -115,7 +120,7 @@ using OrtApiBaseFn = const OrtApiBase* (*)(void); /// Before using this C++ wrapper API, you MUST call Ort::InitApi to set the below 'api' variable inline const OrtApi* api{}; -#if defined(__linux__) +#if defined(__linux__) || defined(MACOS_USE_DLOPEN) inline std::string GetCurrentModuleDir() { Dl_info dl_info; dladdr((void*)GetCurrentModuleDir, &dl_info); @@ -135,14 +140,16 @@ inline void* LoadDynamicLibraryIfExists(const std::string& path) { if (ort_lib_handle == nullptr) { char* err = dlerror(); LOG_WARN("Error while dlopen: %s", (err != nullptr ? err : "Unknown")); - // Trying current dir - std::string current_module_dir = GetCurrentModuleDir(); - std::string local_path{current_module_dir + "/" + path}; - LOG_INFO("Attempting to dlopen %s", local_path.c_str()); - ort_lib_handle = dlopen(local_path.c_str(), RTLD_NOW | RTLD_LOCAL); + if (path.front() != '/') { + // If not absolute path, try search for current dir + std::string current_module_dir = GetCurrentModuleDir(); + std::string local_path{current_module_dir + "/" + path}; + LOG_INFO("Attempting to dlopen %s", local_path.c_str()); + ort_lib_handle = dlopen(local_path.c_str(), RTLD_NOW | RTLD_LOCAL); + } } if (ort_lib_handle) { -#if !defined(__ANDROID__) // RTLD_DI_ORIGIN not available on Android +#if !defined(__ANDROID__) && !defined(__APPLE__) // RTLD_DI_ORIGIN not available on Android & Darwin char pathname[PATH_MAX]; dlinfo((void*)ort_lib_handle, RTLD_DI_ORIGIN, &pathname); LOG_INFO("Loaded native library at %s", pathname); @@ -196,7 +203,7 @@ inline void InitApi() { Generators::SetLogBool("ort_lib", true); } -#if defined(__linux__) +#if defined(__linux__) || defined(MACOS_USE_DLOPEN) // If the GenAI library links against the onnxruntime library, it will have a dependency on a specific // version of OrtGetApiBase. // @@ -218,15 +225,31 @@ inline void InitApi() { // any libonnxruntime.so that supports one of those versions. // - const std::string path = "libonnxruntime.so"; // "libonnxruntime4j_jni.so" is also an option if we have issues - void* ort_lib_handle = LoadDynamicLibraryIfExists(path); + void* ort_lib_handle = nullptr; + const char* ort_lib_path = std::getenv("ORT_LIB_PATH"); + if (ort_lib_path) { + ort_lib_handle = LoadDynamicLibraryIfExists(ort_lib_path); + } -#if !defined(__ANDROID__) +#if defined(__linux__) if (ort_lib_handle == nullptr) { + // For Android and NuGet Linux package, the file name is libonnxruntime.so + // "libonnxruntime4j_jni.so" is also an option on Android if we have issues + ort_lib_handle = LoadDynamicLibraryIfExists("libonnxruntime.so"); + } + + if (ort_lib_handle == nullptr) { + // On Linux it can also be `libonnxruntime.so.1`. See: https://github.com/microsoft/onnxruntime/pull/21339 ort_lib_handle = LoadDynamicLibraryIfExists("libonnxruntime.so.1"); } #endif +#if defined(MACOS_USE_DLOPEN) + if (ort_lib_handle == nullptr) { + ort_lib_handle = LoadDynamicLibraryIfExists("libonnxruntime.dylib"); + } +#endif + if (ort_lib_handle == nullptr) { throw std::runtime_error(std::string("Failed to load onnxruntime. Set ORTGENAI_LOG_ORT_LIB envvar to enable detailed logging.")); } @@ -238,11 +261,11 @@ inline void InitApi() { } InitApiWithDynamicFn(ort_api_base_fn); -#else // defined(__linux__) +#else // defined(__linux__) || defined(MACOS_USE_DLOPEN) api = OrtGetApiBase()->GetApi(ORT_API_VERSION); if (!api) throw std::runtime_error("Onnxruntime is installed but is too old, please install a newer version"); -#endif // defined(__linux__) +#endif // defined(__linux__) || defined(MACOS_USE_DLOPEN) } /** \brief All C++ methods that can fail will throw an exception of this type diff --git a/src/python/CMakeLists.txt b/src/python/CMakeLists.txt index 6628862a3..c17a944ad 100644 --- a/src/python/CMakeLists.txt +++ b/src/python/CMakeLists.txt @@ -9,7 +9,7 @@ target_include_directories(python PRIVATE ${ORT_HEADER_DIR}) target_link_directories(python PRIVATE ${ORT_LIB_DIR}) target_link_libraries(python PRIVATE onnxruntime-genai-static) -if(NOT (CMAKE_SYSTEM_NAME STREQUAL "Android" OR CMAKE_SYSTEM_NAME STREQUAL "Linux")) +if(NOT (CMAKE_SYSTEM_NAME STREQUAL "Android" OR CMAKE_SYSTEM_NAME STREQUAL "Linux" OR CMAKE_SYSTEM_NAME STREQUAL "Darwin")) target_link_libraries(python PRIVATE ${ONNXRUNTIME_LIB}) endif() diff --git a/src/python/py/_dll_directory.py b/src/python/py/_dll_directory.py index 0e47dbdce..1b09994c0 100644 --- a/src/python/py/_dll_directory.py +++ b/src/python/py/_dll_directory.py @@ -12,6 +12,10 @@ def _is_linux(): return sys.platform.startswith("linux") +def _is_macos(): + return sys.platform.startswith("darwin") + + def add_onnxruntime_dependency(package_id: str): """Add the onnxruntime shared library dependency. @@ -38,7 +42,7 @@ def add_onnxruntime_dependency(package_id: str): import ctypes _ = ctypes.CDLL(dml_path) - elif _is_linux(): + elif _is_linux() or _is_macos(): import importlib.util import ctypes import glob @@ -50,11 +54,17 @@ def add_onnxruntime_dependency(package_id: str): # Load the onnxruntime shared library here since we can find the path in python with ease. # This avoids needing to know the exact path of the shared library from native code. ort_package_path = ort_package.submodule_search_locations[0] - ort_lib_path = glob.glob(os.path.join(ort_package_path, "capi", "libonnxruntime.so*")) + if _is_linux(): + ort_lib_path = glob.glob(os.path.join(ort_package_path, "capi", "libonnxruntime.so*")) + elif _is_macos(): + ort_lib_path = glob.glob(os.path.join(ort_package_path, "capi", "libonnxruntime*.dylib")) if not ort_lib_path: raise ImportError("Could not find the onnxruntime shared library.") - _ = ctypes.CDLL(ort_lib_path[0]) + target_lib_path = ort_lib_path[0] + os.environ["ORT_LIB_PATH"] = target_lib_path + + _ = ctypes.CDLL(target_lib_path) def add_cuda_dependency(): diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt index da3502bb4..cb5967171 100644 --- a/test/CMakeLists.txt +++ b/test/CMakeLists.txt @@ -26,7 +26,7 @@ target_link_libraries(unit_tests PRIVATE GTest::gtest_main ) -if(NOT (CMAKE_SYSTEM_NAME STREQUAL "Android" OR CMAKE_SYSTEM_NAME STREQUAL "Linux")) +if(NOT (CMAKE_SYSTEM_NAME STREQUAL "Android" OR CMAKE_SYSTEM_NAME STREQUAL "Linux" OR CMAKE_SYSTEM_NAME STREQUAL "Darwin")) target_link_libraries(unit_tests PRIVATE ${ONNXRUNTIME_LIB}) endif() diff --git a/test/csharp/Microsoft.ML.OnnxRuntimeGenAI.Tests.csproj b/test/csharp/Microsoft.ML.OnnxRuntimeGenAI.Tests.csproj index 7d6c8ea74..711719a34 100644 --- a/test/csharp/Microsoft.ML.OnnxRuntimeGenAI.Tests.csproj +++ b/test/csharp/Microsoft.ML.OnnxRuntimeGenAI.Tests.csproj @@ -63,6 +63,16 @@ false + + + PreserveNewest + false + + + PreserveNewest + false + + PreserveNewest false diff --git a/test/python/requirements-macos.txt b/test/python/requirements-macos.txt new file mode 100644 index 000000000..dcf4c3d97 --- /dev/null +++ b/test/python/requirements-macos.txt @@ -0,0 +1,4 @@ +-f https://download.pytorch.org/whl/torch_stable.html +-i https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/ORT-Nightly/pypi/simple/ +torch==2.2.1 +ort-nightly==1.20.0.dev20240903006 diff --git a/tools/nuget/generate_nuspec_for_native_nuget.py b/tools/nuget/generate_nuspec_for_native_nuget.py index 5ea725e17..ba464add6 100644 --- a/tools/nuget/generate_nuspec_for_native_nuget.py +++ b/tools/nuget/generate_nuspec_for_native_nuget.py @@ -122,7 +122,7 @@ def add_native_artifact_if_exists(xml_lines, runtime, artifact): f'' ) - runtimes = ["win-x64", "win-arm64", "linux-x64"] + runtimes = ["win-x64", "win-arm64", "linux-x64", "osx-x64", "osx-arm64"] for runtime in runtimes: if runtime.startswith("win"): add_native_artifact_if_exists(lines, runtime, "onnxruntime-genai.lib") @@ -130,6 +130,8 @@ def add_native_artifact_if_exists(xml_lines, runtime, artifact): add_native_artifact_if_exists(lines, runtime, "d3d12core.dll") if runtime.startswith("linux"): add_native_artifact_if_exists(lines, runtime, "libonnxruntime-genai.so") + if runtime.startswith("osx"): + add_native_artifact_if_exists(lines, runtime, "libonnxruntime-genai.dylib") # targets for dotnet in ["netstandard2.0", "net8.0", "native"]: