Source Code Compilation and Build

Obtain the vLLM source code, build a CUDA wheel from the source package, and install the build product.

1. Obtain the vLLM 0.18.1 source code.

   1	git clone --branch v0.18.1 --depth 1 https://github.com/vllm-project/vllm.git

2. Build the CUDA wheel from source code.

   1 2 3 4 5 6 7 8 9 10 11 12

   cd vllm export CUDA_HOME=/usr/local/cuda-13.0 export PATH=$CUDA_HOME/bin:$PATH export LD_LIBRARY_PATH=$CUDA_HOME/lib64:${LD_LIBRARY_PATH:-} export VLLM_TARGET_DEVICE=cuda export TORCH_CUDA_ARCH_LIST="8.0" export CMAKE_CUDA_ARCHITECTURES=80 export MAX_JOBS=8 export NVCC_THREADS=1 export CMAKE_BUILD_PARALLEL_LEVEL=8 python3 -m pip wheel --no-build-isolation --no-deps \ --wheel-dir /tmp/vllm-build/dist .

3. Install the local build product.

   1	python3 -m pip install --force-reinstall --no-deps /tmp/vllm-build/dist/vllm-0.18.1+cu130-*.whl

   The following is an example of the build product.

   1	vllm-0.18.1+cu130-cp311-cp311-linux_aarch64.whl