hf_text-generation-inference/server/Makefile-flash-att-v2

30 lines
1.3 KiB
Plaintext

flash_att_v2_commit_cuda := 02ac572f3ffc4f402e4183aaa6824b45859d3ed3
flash_att_v2_commit_rocm := 8736558c287ff2ef28b24878e42828c595ac3e69
flash-attention-v2-cuda:
# Clone flash attention
pip install -U packaging ninja --no-cache-dir
git clone https://github.com/HazyResearch/flash-attention.git flash-attention-v2
build-flash-attention-v2-cuda: flash-attention-v2-cuda
cd flash-attention-v2 && git fetch && git checkout $(flash_att_v2_commit_cuda)
cd flash-attention-v2 && git submodule update --init --recursive
cd flash-attention-v2 && python setup.py build
install-flash-attention-v2-cuda: build-flash-attention-v2-cuda
cd flash-attention-v2 && git submodule update --init --recursive && python setup.py install
flash-attention-v2-rocm:
# Clone flash attention
pip install -U packaging ninja --no-cache-dir
git clone https://github.com/fxmarty/flash-attention-rocm flash-attention-v2
build-flash-attention-v2-rocm: flash-attention-v2-rocm
cd flash-attention-v2 && git fetch && git checkout $(flash_att_v2_commit_rocm)
cd flash-attention-v2 && git submodule update --init --recursive
cd flash-attention-v2 && PYTORCH_ROCM_ARCH=gfx90a python setup.py build
install-flash-attention-v2-rocm: build-flash-attention-v2-rocm
cd flash-attention-v2 && git submodule update --init --recursive && python setup.py install