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

flash_att_v2_commit_cuda := v2.6.1
flash_att_v2_commit_rocm := 2092111b9f975b3347c652ff7fabd431130256c4

build-flash-attention-v2-cuda:
	pip install -U packaging wheel
	pip install flash-attn==$(flash_att_v2_commit_cuda)

install-flash-attention-v2-cuda: build-flash-attention-v2-cuda
	echo "Flash v2 installed"

build-flash-attention-v2-rocm:
	if [ ! -d 'flash-attention-v2' ]; then \
		pip install -U packaging ninja  --no-cache-dir && \
		git clone https://github.com/mht-sharma/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 build; \
	fi

install-flash-attention-v2-rocm: build-flash-attention-v2-rocm
	cd flash-attention-v2 &&  \
	GPU_ARCHS="gfx90a;gfx942" PYTORCH_ROCM_ARCH="gfx90a;gfx942" python setup.py install