feat: CI Add packaging job for cudart runtime

.github/workflows/build.yml

@@ -288,17 +288,6 @@ jobs:
     permissions:
       contents: write
 
-    env:
-      OPENBLAS_VERSION: 0.3.23
-      OPENCL_VERSION: 2023.04.17
-      CLBLAST_VERSION: 1.6.0
-
-    strategy:
-      matrix:
-        include:
-          - build: 'normal'
-            defines: '-DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DLLAMA_BLAS=ON -DBUILD_SHARED_LIBS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS -DBLAS_INCLUDE_DIRS="$env:RUNNER_TEMP/openblas/include" -DBLAS_LIBRARIES="$env:RUNNER_TEMP/openblas/lib/openblas.lib"'
-
     steps:
       - name: Clone
 
@@ -315,7 +304,6 @@ jobs:
         env:
           ACTIONS_ALLOW_UNSECURE_COMMANDS: true
 
-
       - name: actions-setup-cmake
         uses: jwlawson/actions-setup-cmake@v1.14.1
 
@@ -327,15 +315,15 @@ jobs:
           cmake --build ./build_deps/nitro_deps --config Release
           mkdir -p build
           cd build
-          cmake ..
+          cmake .. -DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DLLAMA_BLAS=ON -DBUILD_SHARED_LIBS=ON
           cmake --build . --config Release -j "%NUMBER_OF_PROCESSORS%"
 
       - name: Pack artifacts
         id: pack_artifacts
         shell: cmd
         run: |
-          robocopy build_deps\_install\bin .\build\Release zlib.dll
-          robocopy build\bin\Release .\build\Release llama.dll
+          robocopy build_deps\_install\bin\ .\build\Release\ zlib.dll
+          robocopy build\bin\Release\ .\build\Release\ llama.dll
           dotnet tool install --global AzureSignTool
           azuresigntool.exe sign -kvu "${{ secrets.AZURE_KEY_VAULT_URI }}" -kvi "${{ secrets.AZURE_CLIENT_ID }}" -kvt "${{ secrets.AZURE_TENANT_ID }}" -kvs "${{ secrets.AZURE_CLIENT_SECRET }}" -kvc ${{ secrets.AZURE_CERT_NAME }} -tr http://timestamp.globalsign.com/tsa/r6advanced1 -v ".\build\Release\nitro.exe"
           7z a -ttar temp.tar .\build\Release\*
@@ -371,11 +359,6 @@ jobs:
     permissions:
       contents: write
 
-    strategy:
-      matrix:
-        cuda: ['12.2.0']
-        build: ['cublas']
-
     steps:
       - name: Clone
         id: checkout
@@ -390,7 +373,7 @@ jobs:
           silent: true
         env:
           ACTIONS_ALLOW_UNSECURE_COMMANDS: true
-
+    
       - uses: actions/setup-dotnet@v3
         with:
           dotnet-version: '6.0.x'
@@ -403,16 +386,16 @@ jobs:
           cmake --build ./build_deps/nitro_deps --config Release
           mkdir -p build
           cd build
-          cmake .. -DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DLLAMA_CUBLAS=ON
+          cmake .. -DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DLLAMA_CUBLAS=ON -DBUILD_SHARED_LIBS=ON
           cmake --build . --config Release -j "%NUMBER_OF_PROCESSORS%"
 
       - name: Pack artifacts
         id: pack_artifacts
         shell: cmd
         run: |
           set PATH=%PATH%;C:\Program Files\7-Zip\
-          robocopy build_deps\_install\bin .\build\Release zlib.dll
-          robocopy build\bin\Release .\build\Release llama.dll
+          robocopy build_deps\_install\bin\ .\build\Release\ zlib.dll
+          robocopy build\bin\Release\ .\build\Release\ llama.dll
           dotnet tool install --global AzureSignTool
           %USERPROFILE%\.dotnet\tools\azuresigntool.exe sign -kvu "${{ secrets.AZURE_KEY_VAULT_URI }}" -kvi "${{ secrets.AZURE_CLIENT_ID }}" -kvt "${{ secrets.AZURE_TENANT_ID }}" -kvs "${{ secrets.AZURE_CLIENT_SECRET }}" -kvc ${{ secrets.AZURE_CERT_NAME }} -tr http://timestamp.globalsign.com/tsa/r6advanced1 -v ".\build\Release\nitro.exe"
           7z a -ttar temp.tar .\build\Release\*
@@ -439,11 +422,52 @@ jobs:
           upload_url: ${{ needs.create-draft-release.outputs.upload_url }}
           asset_path: ./nitro.tar.gz
           asset_name: nitro-${{ needs.create-draft-release.outputs.version }}-win-amd64-cuda.tar.gz
-          asset_content_type: application/gzip     
+          asset_content_type: application/gzip
+
+  windows-amd64-cuda-package:
+    runs-on: windows-latest
+    needs: create-draft-release
+    if: always() && (needs.create-draft-release.result == 'success' || needs.create-draft-release.result == 'skipped')
+    permissions:
+      contents: write
+
+    strategy:
+      matrix:
+        cuda: ['12.2.0', '11.7.1']
+        build: ['cublas']
+
+    steps:
+      - uses: Jimver/cuda-toolkit@v0.2.11
+        id: cuda-toolkit
+        with:
+          cuda: ${{ matrix.cuda }}
+          method: 'network'
+          sub-packages: '["nvcc", "cudart", "cublas", "cublas_dev", "thrust", "visual_studio_integration"]'
+
+      - name: Copy and Pack CUDA runtime
+        id: pack_artifacts
+        shell: cmd
+        run: |
+          echo "Cuda install location: ${{steps.cuda-toolkit.outputs.CUDA_PATH}}"
+          mkdir -p build
+          $dst='.\build\'
+          robocopy "${{steps.cuda-toolkit.outputs.CUDA_PATH}}\bin" $dst cudart64_*.dll cublas64_*.dll cublasLt64_*.dll
+          7z a -ttar temp.tar $dst\*
+          7z a -tgzip cudart.tar.gz temp.tar
+
+      - uses: actions/upload-release-asset@v1.0.1
+        if: github.event_name == 'push' && startsWith(github.ref, 'refs/tags/')
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        with:
+          upload_url: ${{ needs.create-draft-release.outputs.upload_url }}
+          asset_path: ./cudart.tar.gz
+          asset_name: cudart-bin-win-cu${{ matrix.cuda }}-x64.tar.gz
+          asset_content_type: application/gzip
 
   update_release_draft:
     if: github.event_name == 'push' && startsWith(github.ref, 'refs/tags/')
-    needs: [ubuntu-amd64-build, ubuntu-amd64-cuda-build, macOS-M-build, macOS-Intel-build, windows-amd64-build, windows-amd64-cuda-build]
+    needs: [ubuntu-amd64-build, ubuntu-amd64-cuda-build, macOS-M-build, macOS-Intel-build, windows-amd64-build, windows-amd64-cuda-build, windows-amd64-cuda-package]
     permissions:
       contents: write
       pull-requests: write