mirror of
https://github.com/huggingface/text-generation-inference.git
synced 2025-09-11 20:34:54 +00:00
Putting back build steps for rocm.
This commit is contained in:
parent
bc925070d3
commit
097f7e9b88
@ -9,10 +9,16 @@ install-flash-attention-v2-cuda:
|
||||
pip install -U packaging wheel
|
||||
pip install flash-attn==$(flash_att_v2_commit_cuda)
|
||||
|
||||
install-flash-attention-v2-rocm: build-flash-attention-v2-rocm
|
||||
build-flash-attention-v2-rocm:
|
||||
if [ ! -d 'flash-attention-v2' ]; then \
|
||||
pip install -U packaging ninja --no-cache-dir && \
|
||||
git clone https://github.com/ROCm/flash-attention.git flash-attention-v2 && \
|
||||
cd flash-attention-v2 && git fetch && git checkout $(flash_att_v2_commit_rocm) && \
|
||||
git submodule update --init --recursive && GPU_ARCHS="gfx90a;gfx942" PYTORCH_ROCM_ARCH="gfx90a;gfx942" python setup.py install; \
|
||||
git submodule update --init --recursive && GPU_ARCHS="gfx90a;gfx942" PYTORCH_ROCM_ARCH="gfx90a;gfx942" python setup.py build; \
|
||||
fi
|
||||
|
||||
install-flash-attention-v2-rocm: build-flash-attention-v2-rocm
|
||||
if [ ! -d 'flash-attention-v2' ]; then \
|
||||
cd flash-attention-v2 && \
|
||||
GPU_ARCHS="gfx90a;gfx942" PYTORCH_ROCM_ARCH="gfx90a;gfx942" python setup.py install; \
|
||||
fi
|
||||
|
@ -11,10 +11,16 @@ install-vllm-cuda: build-vllm-cuda
|
||||
cd vllm && pip install -e .; \
|
||||
fi
|
||||
|
||||
install-vllm-rocm:
|
||||
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 && \
|
||||
cd vllm && git fetch && git checkout ca6913b3c2ffacdcb7d15e914dc34adbc6c89479 && \
|
||||
PYTORCH_ROCM_ARCH="gfx90a;gfx942" python setup.py build; \
|
||||
fi
|
||||
|
||||
install-vllm-rocm: build-vllm-rocm
|
||||
if [ ! -d 'vllm' ]; then \
|
||||
cd vllm && \
|
||||
PYTORCH_ROCM_ARCH="gfx90a;gfx942" pip install -e .; \
|
||||
fi
|
||||
|
Loading…
Reference in New Issue
Block a user