hf_text-generation-inference/server/Makefile-vllm

24 lines
896 B
Plaintext

commit_cuda := d243e9dc7e2c9c2e36a4150ec8e64809cb55c01b
commit_rocm := 4e0929e6e4fa0a3d09d358715c288020ea9dc247
build-vllm-cuda:
if [ ! -d 'vllm' ]; then \
pip install -U ninja packaging --no-cache-dir && \
git clone https://github.com/Narsil/vllm.git vllm; \
fi
cd vllm && git fetch origin && git checkout $(commit_cuda) && python setup.py build
install-vllm-cuda: build-vllm-cuda
cd vllm && git fetch origin && git checkout $(commit_cuda) && pip install -e .
build-vllm-rocm:
if [ ! -d 'vllm' ]; then \
pip install -U ninja packaging --no-cache-dir && \
git clone https://github.com/mht-sharma/vllm.git vllm; \
fi
cd vllm && git fetch && git checkout $(commit_rocm) && \
PYTORCH_ROCM_ARCH="gfx90a;gfx942" python setup.py build
install-vllm-rocm: build-vllm-rocm
cd vllm && git fetch && git checkout $(commit_rocm) && \
PYTORCH_ROCM_ARCH="gfx90a;gfx942" pip install -e .