mirror of
https://github.com/huggingface/text-generation-inference.git
synced 2025-04-22 15:32:08 +00:00
6aeb669072
* Softcapping for gemma2. * Less clutter. * No access to transformers config, only config_dict here. * 0.0 is the null value in the C++ API.
22 lines
918 B
Plaintext
22 lines
918 B
Plaintext
flash_att_v2_commit_cuda := v2.6.1
|
|
flash_att_v2_commit_rocm := 2554f490101742ccdc56620a938f847f61754be6
|
|
|
|
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/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 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
|