huggingface/text-generation-inference
5
0
Fork 0
mirror of https://github.com/huggingface/text-generation-inference.git synced 2025-09-11 04:14:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

make install-flash-attn-v2-cuda should work like `make

Browse Source 
install-flash-attn-v2` used to work.
This commit is contained in:
Nicolas Patry 2023-11-28 15:03:08 +01:00
parent b2b5df0e94
commit 1c09961f14
1 changed files with 22 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

39
server/Makefile-flash-att-v2
View File

@ -1,26 +1,29 @@
flash_att_v2_commit := 02ac572f3ffc4f402e4183aaa6824b45859d3ed3 flash_att_v2_commit_cuda := 02ac572f3ffc4f402e4183aaa6824b45859d3ed3
flash_att_v2_commit_rocm := 8736558c287ff2ef28b24878e42828c595ac3e69
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 flash-attention-v2-cuda:
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 # Clone flash attention
pip install -U packaging ninja --no-cache-dir pip install -U packaging ninja --no-cache-dir
git clone --single-branch --branch $(BRANCH) $(FLASH_REPOSITORY) flash-attention-v2 git clone https://github.com/HazyResearch/flash-attention.git flash-attention-v2
build-flash-attention-v2: flash-attention-v2 build-flash-attention-v2-cuda: flash-attention-v2-cuda
cd flash-attention-v2 && git fetch && git checkout $(FLASH_ATTN_V2_COMMIT) cd flash-attention-v2 && git fetch && git checkout $(flash_att_v2_commit_cuda)
cd flash-attention-v2 && git submodule update --init --recursive cd flash-attention-v2 && git submodule update --init --recursive
cd flash-attention-v2 && PYTORCH_ROCM_ARCH=$(PYTORCH_ROCM_ARCH) python setup.py build cd flash-attention-v2 && python setup.py build
install-flash-attention-v2: build-flash-attention-v2 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 cd flash-attention-v2 && git submodule update --init --recursive && python setup.py install