hf_text-generation-inference/server/Makefile-vllm

24 lines
885 B
Plaintext

commit_cuda := b5dfc61db88a81069e45b44f7cc99bd9e62a60fa
commit_rocm := ca6913b3c2ffacdcb7d15e914dc34adbc6c89479
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 && git checkout $(commit_cuda) && python setup.py build
install-vllm-cuda: build-vllm-cuda
cd vllm && git fetch && 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/fxmarty/rocm-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 .