Add SYCL backend support

.github/workflows/build-aar.yml

@@ -1,11 +1,13 @@
 name: Build Android AAR
 
 on:
-  push:
-    branches:
-      - main
-      - android-sdk
-  pull_request:
+  workflow_dispatch:
+  # AAR pre-build task temporarily disabled
+  # push:
+  #   branches:
+  #     - main
+  #     - android-sdk
+  # pull_request:
 
 jobs:
   build:
@@ -101,4 +103,4 @@ jobs:
         upload_url: ${{ steps.create_release.outputs.upload_url }}
         asset_path: release-artifacts/com-nexa-${{ github.ref_name }}-${{ github.run_number }}.aar
         asset_name: com-nexa-${{ github.ref_name }}-${{ github.run_number }}.aar
-        asset_content_type: application/java-archive
+        asset_content_type: application/java-archive

.github/workflows/build-wheels-cpu.yaml

@@ -6,48 +6,6 @@ permissions:
   contents: write
 
 jobs:
-  # build_wheels_linux:
-  #   name: Build wheels on ${{ matrix.os }}
-  #   runs-on: ${{ matrix.os }}
-  #   strategy:
-  #     matrix:
-  #       os: [ubuntu-20.04]
-
-  #   steps:
-  #     - uses: actions/checkout@v4
-  #       with:
-  #         submodules: "recursive"
-
-  #     # Used to host cibuildwheel
-  #     - uses: actions/setup-python@v5
-  #       with:
-  #         python-version: "3.8"
-  #         cache: "pip"
-
-  #     - name: Install dependencies
-  #       run: |
-  #         python -m pip install --upgrade pip
-  #         # python -m pip install -e .
-  #         python -m pip install build wheel
-
-  #     - name: Build wheels
-  #       uses: pypa/[email protected]
-  #       env:
-  #         # disable repair
-  #         CIBW_REPAIR_WHEEL_COMMAND: ""
-  #         CIBW_BUILD_FRONTEND: "build"
-  #         CIBW_SKIP: "*musllinux*"
-  #         CIBW_BUILD: "cp37-* cp38-* cp39-* cp310-* cp311-* cp312-*"
-  #         CMAKE_BUILD_PARALLEL_LEVEL: 4
-  #       with:
-  #         package-dir: .
-  #         output-dir: wheelhouse
-
-  #     - uses: actions/upload-artifact@v4
-  #       with:
-  #         name: wheels-${{ matrix.os }}
-  #         path: ./wheelhouse/*.whl
-
   build_wheels_win:
     name: Build wheels on ${{ matrix.os }}
     runs-on: ${{ matrix.os }}
@@ -100,14 +58,6 @@ jobs:
         with:
           submodules: "recursive"
 
-      # - name: Set up QEMU
-      #   uses: docker/setup-qemu-action@v3
-      #   with:
-      #     platforms: linux/arm64
-
-      # - name: Check nproc value
-      #   run: echo "Number of processors $(nproc)"
-
       - name: Build wheels
         uses: pypa/[email protected]
         env:

.github/workflows/build-wheels-cuda-linux.yaml

@@ -71,15 +71,6 @@ jobs:
           $cudaVersion = $env:CUDAVER
           mamba install -y 'cuda' -c nvidia/label/cuda-$cudaVersion
           python -m pip install build wheel
-
-      # - name: Debug GLIBC Issues
-      #   run: |
-      #     echo "Check GCC"
-      #     gcc --version
-      #     which gcc
-
-      #     echo "## Checking if libdl and librt are available"
-      #     ldconfig -p | grep -E "libdl|librt" || echo "Missing libdl or librt!"
 
       - name: Build Wheel
         run: |
@@ -93,7 +84,7 @@ jobs:
           $env:VERBOSE = '1'
 
           $env:CMAKE_BUILD_PARALLEL_LEVEL = $(nproc)
-          $env:CMAKE_ARGS = '-DSD_CUBLAS=ON -DGGML_CUDA=ON -DGGML_CUBLAS=ON -DCMAKE_CUDA_ARCHITECTURES=all'
+          $env:CMAKE_ARGS = '-DSD_CUDA=ON -DGGML_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES=all'
           $env:CMAKE_ARGS = $env:CMAKE_ARGS + ' -DGGML_CUDA_FORCE_MMQ=ON -DGGML_AVX2=OFF -DGGML_FMA=OFF -DGGML_F16C=OFF'
 
           python -m build --wheel
@@ -106,4 +97,4 @@ jobs:
           # Set tag_name to <tag>-cu<cuda_version>
           tag_name: ${{ github.ref_name }}-cu${{ env.CUDA_VERSION }}
         env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/build-wheels-cuda-win.yaml

@@ -115,7 +115,7 @@ jobs:
           }
           $env:VERBOSE = '1'
 
-          $env:CMAKE_ARGS = '-DSD_CUBLAS=ON -DGGML_CUDA=ON -DGGML_CUBLAS=ON -DCMAKE_CUDA_ARCHITECTURES=all'
+          $env:CMAKE_ARGS = '-DSD_CUDA=ON -DGGML_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES=all'
           $env:CMAKE_ARGS = $env:CMAKE_ARGS + ' -DGGML_CUDA_FORCE_MMQ=ON -DGGML_AVX2=OFF -DGGML_FMA=OFF -DGGML_F16C=OFF'
           $env:CMAKE_ARGS = $env:CMAKE_ARGS + ' -DCMAKE_SYSTEM_VERSION=10.0.18362.0 -DCMAKE_VS_WINDOWS_TARGET_PLATFORM_VERSION=10.0.18362.0'
           $env:CMAKE_BUILD_PARALLEL_LEVEL = $(nproc)

.github/workflows/build-wheels-rocm-linux.yaml

@@ -86,4 +86,4 @@ jobs:
           # Set tag_name to <tag>-rocm<rocm_version>
           tag_name: ${{ github.ref_name }}-rocm${{ env.ROCM_VERSION }}
         env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/build-wheels-sycl-win.yaml

@@ -0,0 +1,100 @@
+name: Build Wheels (SYCL) (Windows)
+on:
+  workflow_dispatch:
+
+permissions:
+  contents: write
+
+jobs:
+  define_matrix:
+    name: Define Build Matrix
+    runs-on: sdk-windows-8-core
+    outputs:
+      matrix: ${{ steps.set-matrix.outputs.matrix }}
+    defaults:
+      run:
+        shell: pwsh
+    steps:
+      - name: Define Job Output
+        id: set-matrix
+        run: |
+          $matrix = @{
+              'os' = @('sdk-windows-8-core')
+              'pyver' = @("3.8", "3.9", "3.10", "3.11", "3.12")
+              'releasetag' = @("basic")
+          }
+          $matrixOut = ConvertTo-Json $matrix -Compress
+          Write-Output ('matrix=' + $matrixOut) >> $env:GITHUB_OUTPUT
+
+  build_wheels:
+    name: Build Wheel ${{ matrix.os }} Python ${{ matrix.pyver }}
+    needs: define_matrix
+    runs-on: ${{ matrix.os }}
+    strategy:
+      matrix: ${{ fromJSON(needs.define_matrix.outputs.matrix) }}
+    defaults:
+      run:
+        shell: pwsh
+    env:
+      WINDOWS_BASEKIT_URL: https://registrationcenter-download.intel.com/akdlm/IRC_NAS/b380d914-366b-4b77-a74a-05e3c38b3514/intel-oneapi-base-toolkit-2025.0.0.882_offline.exe
+      WINDOWS_DPCPP_MKL: intel.oneapi.win.cpp-dpcpp-common:intel.oneapi.win.mkl.devel:intel.oneapi.win.dnnl:intel.oneapi.win.tbb.devel
+    steps:
+      - name: Add MSBuild to PATH
+        if: runner.os == 'Windows'
+        uses: microsoft/setup-msbuild@v2
+        with:
+          vs-version: "[16.11,16.12)"
+          msbuild-architecture: x64
+
+      - name: Checkout Repository
+        uses: actions/checkout@v4
+        with:
+          submodules: "recursive"
+
+      - name: Setup Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.pyver }}
+          architecture: 'x64'
+          cache: "pip"
+
+      - name: Install Ninja Build System
+        run: |
+            choco install ninja -y
+            # choco install wget -y
+
+      - name: Install SYCL dependencies
+        run:  |
+          scripts/install-oneapi.bat "${{ env.WINDOWS_BASEKIT_URL }}" "${{ env.WINDOWS_DPCPP_MKL }}"
+
+      - name: Install Python Dependencies
+        run: |
+          python -m pip install --upgrade pip
+          python -m pip install build wheel setuptools cmake
+          python -m pip install scikit-build
+
+      - name: Build Wheel
+        run: |
+          # Add choco bin path to PATH, necessary for using Ninja
+          $env:PATH="C:\ProgramData\chocolatey\bin;$env:PATH"	
+
+          # On Windows, SYCL must be built with Ninja and icx compiler
+          # $env:CMAKE_ARGS="-DGGML_SYCL=ON -DCMAKE_C_COMPILER=cl -DCMAKE_CXX_COMPILER=icx"
+
+          # Build the wheel
+          # python -m build --wheel
+          scripts/workflow-win-build-sycl.bat
+
+      - name: Upload Wheel Artifact
+        uses: actions/upload-artifact@v4
+        with:
+          path: dist/*.whl
+          name: llama-sycl-wheel-python${{ matrix.pyver }}.whl
+
+      - name: Create GitHub Release
+        uses: softprops/action-gh-release@v2
+        with:
+          files: dist/*
+          tag_name: ${{ github.ref_name }}-sycl
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/build-wheels-vulkan-win.yaml

@@ -90,4 +90,4 @@ jobs:
           files: dist/*
           tag_name: ${{ github.ref_name }}-vulkan
         env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/ci.yaml

@@ -60,4 +60,4 @@ jobs:
     - name: Run tests
       run: |
         python -m pytest tests
-      shell: bash
+      shell: bash