test(ci): build CUDA 11.8 wheels

Author: abetlen

.github/workflows/test-cuda-118-wheels.yaml

@@ -0,0 +1,136 @@
+name: Test CUDA 11.8 Wheels
+
+on:
+  push:
+    branches:
+      - test/cuda-118-wheels-build
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+jobs:
+  build:
+    name: Build CUDA 11.8 wheel on ${{ matrix.os }}
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os:
+          - ubuntu-22.04
+          - windows-2022
+    defaults:
+      run:
+        shell: pwsh
+    env:
+      CUDAVER: 11.8.0
+
+    steps:
+      - name: Set up MSVC
+        if: runner.os == 'Windows'
+        uses: ilammy/msvc-dev-cmd@v1
+        with:
+          arch: x64
+
+      - uses: actions/checkout@v4
+        with:
+          submodules: recursive
+
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.9"
+          cache: pip
+
+      - name: Setup Mamba
+        uses: conda-incubator/setup-miniconda@v3.1.0
+        with:
+          activate-environment: llamacpp
+          python-version: "3.9"
+          miniforge-version: latest
+          add-pip-as-python-dependency: true
+          auto-activate-base: false
+
+      - name: Install dependencies
+        env:
+          MAMBA_DOWNLOAD_FAILFAST: "0"
+          MAMBA_NO_LOW_SPEED_LIMIT: "1"
+        run: |
+          $cudaVersion = $env:CUDAVER
+          $cudaChannel = "nvidia/label/cuda-$cudaVersion"
+          if ($IsLinux) {
+            mamba install -y --channel-priority flexible --override-channels -c $cudaChannel "${cudaChannel}::cuda-toolkit=$cudaVersion" "${cudaChannel}::cuda-nvcc_linux-64" "${cudaChannel}::cuda-cccl" "${cudaChannel}::cuda-cudart" "${cudaChannel}::cuda-cudart-dev"
+            python -m pip install build wheel
+          } elseif ($IsWindows) {
+            mamba install -y --channel-priority flexible --override-channels -c $cudaChannel "${cudaChannel}::cuda-toolkit=$cudaVersion" "${cudaChannel}::cuda-nvcc_win-64" "${cudaChannel}::cuda-cccl" "${cudaChannel}::cuda-cudart" "${cudaChannel}::cuda-cudart-dev"
+            python -m pip install build wheel ninja
+          } else {
+            throw 'Unsupported CUDA wheel build platform'
+          }
+
+      - name: Build wheel
+        run: |
+          $pathSeparator = if ($IsWindows) { ';' } else { ':' }
+          if ($IsWindows) {
+            $cudaRoot = Join-Path $env:CONDA_PREFIX 'Library'
+          } elseif (Test-Path (Join-Path $env:CONDA_PREFIX 'targets/x86_64-linux/include/cuda_runtime.h')) {
+            $cudaRoot = Join-Path $env:CONDA_PREFIX 'targets/x86_64-linux'
+          } else {
+            $cudaRoot = $env:CONDA_PREFIX
+          }
+
+          $env:CUDA_PATH = $cudaRoot
+          $env:CUDA_HOME = $cudaRoot
+          $env:CUDAToolkit_ROOT = $cudaRoot
+          $env:CUDA_TOOLKIT_ROOT_DIR = $cudaRoot
+          $cudaRootCmake = $cudaRoot.Replace('\', '/')
+          $env:CMAKE_ARGS = "-DCUDAToolkit_ROOT=$cudaRootCmake -DCUDA_TOOLKIT_ROOT_DIR=$cudaRootCmake"
+
+          if ($IsLinux) {
+            if (Test-Path '/usr/bin/g++-12') {
+              $env:CC = '/usr/bin/gcc-12'
+              $env:CXX = '/usr/bin/g++-12'
+              $env:CUDAHOSTCXX = '/usr/bin/g++-12'
+              $env:CMAKE_ARGS = "$env:CMAKE_ARGS -DCMAKE_CUDA_HOST_COMPILER=$env:CUDAHOSTCXX"
+            }
+            $env:CPATH = "$cudaRoot/include$pathSeparator$env:CPATH"
+            $env:CPLUS_INCLUDE_PATH = "$cudaRoot/include$pathSeparator$env:CPLUS_INCLUDE_PATH"
+            $env:LIBRARY_PATH = "$cudaRoot/lib$pathSeparator$env:CONDA_PREFIX/lib$pathSeparator$env:LIBRARY_PATH"
+            $env:LD_LIBRARY_PATH = "$cudaRoot/lib$pathSeparator$env:CONDA_PREFIX/lib$pathSeparator$env:LD_LIBRARY_PATH"
+          } else {
+            $ninjaPath = ((Get-Command ninja -ErrorAction Stop).Source).Replace('\', '/')
+            $env:CMAKE_GENERATOR = 'Ninja'
+            $env:CMAKE_MAKE_PROGRAM = $ninjaPath
+            $env:PATH = "$(Join-Path $cudaRoot 'bin')$pathSeparator$env:PATH"
+          }
+
+          if ($IsWindows) {
+            $nvccCandidates = @(
+              (Join-Path $cudaRoot 'bin\nvcc.exe'),
+              (Join-Path $env:CONDA_PREFIX 'Library\bin\nvcc.exe'),
+              (Join-Path $env:CONDA_PREFIX 'bin\nvcc.exe')
+            )
+          } else {
+            $nvccCandidates = @(
+              (Join-Path $env:CONDA_PREFIX 'bin/nvcc'),
+              (Join-Path $env:CONDA_PREFIX 'targets/x86_64-linux/bin/nvcc')
+            )
+          }
+          $nvccPath = $nvccCandidates | Where-Object { Test-Path $_ } | Select-Object -First 1
+          if (-not $nvccPath) {
+            throw 'Failed to find nvcc in the conda environment'
+          }
+
+          $env:CUDACXX = $nvccPath
+          $env:PATH = "$(Split-Path $nvccPath)$pathSeparator$env:PATH"
+          if ($IsWindows) {
+            $nvccPathCmake = $nvccPath.Replace('\', '/')
+            $env:CUDACXX = $nvccPathCmake
+            $env:CMAKE_ARGS = "-DCMAKE_CUDA_COMPILER=$nvccPathCmake -DCMAKE_CUDA_COMPILER_ARG1=-allow-unsupported-compiler -DCMAKE_MAKE_PROGRAM=$env:CMAKE_MAKE_PROGRAM $env:CMAKE_ARGS"
+          }
+
+          $env:VERBOSE = '1'
+          $env:CMAKE_ARGS = "-DGGML_CUDA_FORCE_MMQ=ON -DGGML_CUDA=on -DCMAKE_CUDA_ARCHITECTURES=60-real;61-real;70-real;75-real;80-real;86-real;89-real;90-real;90-virtual -DCMAKE_CUDA_FLAGS=-allow-unsupported-compiler -DCMAKE_CUDA_FLAGS_INIT=-allow-unsupported-compiler $env:CMAKE_ARGS"
+          $env:CMAKE_ARGS = "$env:CMAKE_ARGS -DGGML_AVX2=off -DGGML_FMA=off -DGGML_F16C=off"
+          & $nvccPath --version
+          python -m build --wheel
+          Get-ChildItem dist | Select-Object Name, Length