rllm-cuda.yml

name: rLLM with CUDA

on:
  push:
    branches: [ "disabled-main" ]
  pull_request:
    branches: [ "disabled-main" ]

env:
  CARGO_TERM_COLOR: always
  TORCH_CUDA_ARCH_LIST: 8.0
  CUDA_COMPUTE_CAP: 80
  LIBTORCH_USE_PYTORCH: 1
  LIBTORCH_BYPASS_VERSION_CHECK: 1

jobs:
  build:

    runs-on: ubuntu-latest

    steps:
      - uses: actions/checkout@v3
        with:
          submodules: true
      - run: sudo apt-get install ccache

      - run: sudo df -h
      - run: sudo rm -rf /usr/share/dotnet /opt/ghc /usr/local/share/boost
      - run: sudo df -h

      - run: pip install torch==2.1.0
      - run: sudo df -h

      - uses: Jimver/cuda-toolkit@v0.2.13
        id: cuda-toolkit
        with:
          cuda: '12.3.2'
      
      - run: echo "Installed cuda version is ${{ steps.cuda-toolkit.outputs.cuda }}"
      - run: echo "Cuda install location ${{ steps.cuda-toolkit.outputs.CUDA_PATH }}"

      - run: nvcc -V

      - name: Build rLLM
        run: cargo build --verbose --release
        working-directory: rllm/rllm-cuda

      - run: strip target/release/rllm-cuda
      - name: Artifact upload
        uses: actions/upload-artifact@v4
        with:
          name: rllm-cuda
          path: target/release/rllm-cuda