Merge pull request #331 from janhq/feat/add-windows-cuda-13

test: cuda13 windows llamacpp build

Author: Minh141120

.github/runners/window-13.dockerfile

@@ -0,0 +1,53 @@
+FROM mcr.microsoft.com/windows/server:ltsc2022
+
+SHELL ["powershell", "-Command", "$ErrorActionPreference = 'Stop';$ProgressPreference='silentlyContinue';"]
+
+ARG RUNNER_VERSION=2.311.0
+
+RUN Invoke-WebRequest \
+      -Uri 'https://aka.ms/install-powershell.ps1' \
+      -OutFile install-powershell.ps1; \
+    powershell -ExecutionPolicy Unrestricted -File ./install-powershell.ps1 -AddToPath
+
+RUN Invoke-WebRequest \
+      -Uri https://github.com/actions/runner/releases/download/v$env:RUNNER_VERSION/actions-runner-win-x64-$env:RUNNER_VERSION.zip \
+      -OutFile runner.zip; \
+    Expand-Archive -Path C:/runner.zip -DestinationPath C:/actions-runner; \
+    Remove-Item -Path C:\runner.zip; \
+    setx /M PATH $(${Env:PATH} + \";${Env:ProgramFiles}\Git\bin\")
+
+# Install Chocolatey
+RUN Set-ExecutionPolicy Bypass -Scope Process -Force; \
+    [System.Net.ServicePointManager]::SecurityProtocol = [System.Net.ServicePointManager]::SecurityProtocol -bor 3072; \
+    Invoke-Expression ((New-Object System.Net.WebClient).DownloadString('https://chocolatey.org/install.ps1'))
+
+# Upgrade Git
+RUN choco install git -y; \
+    git --version
+
+# Install 7zip
+RUN choco install 7zip -y; \
+    7z --help
+
+# Install cmake and add to path
+RUN choco install cmake.install -y --installargs '"ADD_CMAKE_TO_PATH=System"'
+
+RUN cmake --version
+
+# Install MSBuild and add to path
+RUN choco install visualstudio2022buildtools -y --package-parameters '"--add Microsoft.VisualStudio.Component.VC.Tools.x86.x64 --add Microsoft.VisualStudio.Component.Windows10SDK.19041 --add Microsoft.VisualStudio.Component.VC.14.29.CLI.Support --add Microsoft.VisualStudio.Component.VC.14.29.CMake.Project --add Microsoft.VisualStudio.Component.VC.14.29.MFC --add Microsoft.VisualStudio.Component.VC.14.29.MSBuild --add Microsoft.VisualStudio.Component.VC.14.29.VCRedist.x86.x64 --add Microsoft.VisualStudio.Component.VC.14.29.VSTools.x86.x64 --add Microsoft.VisualStudio.Component.VC.14.29.VC.ASAN --add Microsoft.VisualStudio.Component.VC.14.29.VC.CLI.Support --add Microsoft.VisualStudio.Component.VC.14.29.VC.CMake.Project --add Microsoft.VisualStudio.Component.VC.14.29.VC.FxCop --add Microsoft.VisualStudio.Component.VC.14.29.VC.MFC --add Microsoft.VisualStudio.Component.VC.14.29.VC.MSBuild --add Microsoft.VisualStudio.Component.VC.14.29.VC.Redist.14.Latest --add Microsoft.VisualStudio.Component.VC.14.29.VC.Tools.x86.x64 --add Microsoft.VisualStudio.Component.VC.14.29.VC.VCUnitTest --add Microsoft.VisualStudio.Component.VC.14.29.VC.x86.x64 --add Microsoft.VisualStudio.Component.VC.14.29.VC.x86.x64.Latest --add Microsoft.VisualStudio.Component.VC.14.29.VisualStudioCppSDK --add Microsoft.VisualStudio.ComponentGroup.NativeDesktop.Win81 --add Microsoft.VisualStudio.ComponentGroup.NativeDesktop.Win10 --add Microsoft.VisualStudio.Component.Windows10SDK.19041 --add Microsoft.VisualStudio.ComponentGroup.NativeDesktop.LegacyBuildTools --add Microsoft.VisualStudio.Component.VC.ATL --add Microsoft.VisualStudio.Component.VC.CLI.Support --add Microsoft.VisualStudio.Component.VC.CoreIde --add Microsoft.VisualStudio.Component.VC.DiagnosticTools --add Microsoft.VisualStudio.Component.VC.Llvm.Clang --add Microsoft.VisualStudio.Component.VC.Llvm.ClangToolset --add Microsoft.VisualStudio.Component.VC.Llvm.Cmake --add Microsoft.VisualStudio.Component.VC.Llvm.Llvm --add Microsoft.VisualStudio.Component.VC.Llvm.Toolset --add Microsoft.VisualStudio.Component.VC.MFC --add Microsoft.VisualStudio.Component.VC.MSBuild --add Microsoft.VisualStudio"'
+
+RUN choco install gzip -y;
+
+# Install CUDA Toolkit 13.0.2 directly from NVIDIA
+RUN Invoke-WebRequest -Uri 'https://developer.download.nvidia.com/compute/cuda/13.0.2/local_installers/cuda_13.0.2_windows.exe' -OutFile cuda_installer.exe; \
+    Start-Process -FilePath .\cuda_installer.exe -ArgumentList '-s' -Wait -NoNewWindow; \
+    Remove-Item -Path .\cuda_installer.exe
+
+# Copy integrated tools to MSBuild
+RUN Copy-Item -Path 'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.0\extras\visual_studio_integration\MSBuildExtensions\*' -Destination 'C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\MSBuild\Microsoft\VC\v170\BuildCustomizations'
+
+ADD runner.ps1 C:/runner.ps1
+SHELL ["powershell", "-Command", "$ErrorActionPreference = 'Stop'; $ProgressPreference = 'SilentlyContinue';"]
+ENTRYPOINT ["C:\\Program Files (x86)\\Microsoft Visual Studio\\2022\\BuildTools\\Common7\\Tools\\VsDevCmd.bat", "-arch=amd64", "&&", "powershell.exe", "-NoLogo", "-ExecutionPolicy", "Bypass"]
+CMD ["powershell.exe", "-ExecutionPolicy", "Unrestricted", "-File", ".\\runner.ps1"]

.github/workflows/menlo-build.yml

@@ -56,7 +56,7 @@ jobs:
 
   build-and-test:
     runs-on: ${{ matrix.runs-on }}
-    needs: [create-draft-release]
+    # needs: [create-draft-release]
     timeout-minutes: 270
     strategy:
       fail-fast: false
@@ -342,6 +342,14 @@ jobs:
             vulkan: false
             ccache: true
             ccache-dir: 'C:\Users\ContainerAdministrator\AppData\Local\ccache'
+          - os: "win"
+            name: "cuda-13-common_cpus-x64"
+            runs-on: "windows-cuda-13-0"
+            cmake-flags: "-DLLAMA_CURL=OFF -DGGML_CUDA=ON -DGGML_NATIVE=OFF -DCMAKE_CUDA_ARCHITECTURES='75;80;86;87;89;120;121' -DGGML_BACKEND_DL=ON -DGGML_CPU_ALL_VARIANTS=ON  -DBUILD_SHARED_LIBS=ON -DCMAKE_CXX_COMPILER_LAUNCHER=ccache -DCMAKE_C_COMPILER_LAUNCHER=ccache -DCMAKE_BUILD_TYPE='Release' -GNinja"
+            run-e2e: false
+            vulkan: false
+            ccache: true
+            ccache-dir: 'C:\Users\ContainerAdministrator\AppData\Local\ccache'
           - os: "win"
             name: "vulkan-common_cpus-x64"
             runs-on: "windows-cuda-11-7"
@@ -592,6 +600,7 @@ jobs:
           wget https://minio.menlo.ai:9000/cicd/dist/cuda-dependencies/13.0/linux/cuda.tar.gz -O /tmp/cudart-llama-bin-linux-cu13.0-x64.tar.gz
           wget https://minio.menlo.ai:9000/cicd/dist/cuda-dependencies/12.0/linux/cuda.tar.gz -O /tmp/cudart-llama-bin-linux-cu12.0-x64.tar.gz
           wget https://minio.menlo.ai:9000/cicd/dist/cuda-dependencies/11.7/linux/cuda.tar.gz -O /tmp/cudart-llama-bin-linux-cu11.7-x64.tar.gz
+          wget https://minio.menlo.ai:9000/cicd/dist/cuda-dependencies/13.0/windows/cuda.tar.gz -O /tmp/cudart-llama-bin-win-cu13.0-x64.tar.gz
           wget https://minio.menlo.ai:9000/cicd/dist/cuda-dependencies/12.0/windows/cuda.tar.gz -O /tmp/cudart-llama-bin-win-cu12.0-x64.tar.gz
           wget https://minio.menlo.ai:9000/cicd/dist/cuda-dependencies/11.7/windows/cuda.tar.gz -O /tmp/cudart-llama-bin-win-cu11.7-x64.tar.gz
 
@@ -634,6 +643,12 @@ jobs:
           echo "  sha512: >-" >> checksum.yml
           echo "    $(sha512sum /tmp/cudart-llama-bin-win-cu12.0-x64.tar.gz | awk '{ print $1 }')" >> checksum.yml
           echo "  size: $(stat -c%s /tmp/cudart-llama-bin-win-cu12.0-x64.tar.gz)" >> checksum.yml
+
+          echo "- url: cudart-llama-bin-win-cu13.0-x64.tar.gz" >> checksum.yml
+          echo "  sha512: >-" >> checksum.yml
+          echo "    $(sha512sum /tmp/cudart-llama-bin-win-cu13.0-x64.tar.gz | awk '{ print $1 }')" >> checksum.yml
+          echo "  size: $(stat -c%s /tmp/cudart-llama-bin-win-cu13.0-x64.tar.gz)" >> checksum.yml
+
           cat checksum.yml
 
       - name: Upload checksum.yml to GitHub Release
@@ -680,6 +695,17 @@ jobs:
           asset_name: cudart-llama-bin-linux-cu11.7-x64.tar.gz
           asset_content_type: application/gzip
 
+      - name: upload cudart-llama-bin-win-cu13.0-x64.tar.gz to Github Release
+        uses: actions/upload-release-asset@v1
+        if: github.event_name == 'push' && startsWith(github.ref, 'refs/tags/')
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        with:
+          upload_url: ${{ needs.create-draft-release.outputs.upload_url }}
+          asset_path: /tmp/cudart-llama-bin-win-cu13.0-x64.tar.gz
+          asset_name: cudart-llama-bin-win-cu13.0-x64.tar.gz
+          asset_content_type: application/gzip
+
       - name: upload cudart-llama-bin-win-cu12.0-x64.tar.gz to Github Release
         uses: actions/upload-release-asset@v1
         if: github.event_name == 'push' && startsWith(github.ref, 'refs/tags/')