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

flash_att_v2_commit := 02ac572f3ffc4f402e4183aaa6824b45859d3ed3

build-flash-attention-v2-cuda: FLASH_ATTN_V2_COMMIT=02ac572f3ffc4f402e4183aaa6824b45859d3ed3
build-flash-attention-v2-cuda: FLASH_REPOSITORY=https://github.com/HazyResearch/flash-attention.git
build-flash-attention-v2-cuda: BRANCH=main
build-flash-attention-v2-cuda: PYTORCH_ROCM_ARCH=""
build-flash-attention-v2-cuda: build-flash-attention-v2

build-flash-attention-v2-rocm: FLASH_ATTN_V2_COMMIT=8736558c287ff2ef28b24878e42828c595ac3e69
build-flash-attention-v2-rocm: FLASH_REPOSITORY=https://github.com/fxmarty/flash-attention-rocm
build-flash-attention-v2-rocm: BRANCH=remove-offload-arch-native
build-flash-attention-v2-rocm: PYTORCH_ROCM_ARCH=gfx90a
build-flash-attention-v2-rocm: build-flash-attention-v2

flash-attention-v2:
  # Clone flash attention
	pip install -U packaging ninja  --no-cache-dir
	git clone --single-branch --branch $(BRANCH) $(FLASH_REPOSITORY) flash-attention-v2

build-flash-attention-v2: flash-attention-v2
	cd flash-attention-v2 && git fetch && git checkout $(FLASH_ATTN_V2_COMMIT)
	cd flash-attention-v2 && git submodule update --init --recursive
	cd flash-attention-v2 && PYTORCH_ROCM_ARCH=$(PYTORCH_ROCM_ARCH) python setup.py build

install-flash-attention-v2: build-flash-attention-v2
	cd flash-attention-v2 && git submodule update --init --recursive && python setup.py install