From bd34d0b86182cb1aa983a849c206311270886e8d Mon Sep 17 00:00:00 2001 From: Wing Lian Date: Sat, 17 May 2025 09:49:26 -0400 Subject: [PATCH] install for hopper from pre-built wheel --- docker/Dockerfile-base | 10 +++------- 1 file changed, 3 insertions(+), 7 deletions(-) diff --git a/docker/Dockerfile-base b/docker/Dockerfile-base index e29e9da12..166002547 100644 --- a/docker/Dockerfile-base +++ b/docker/Dockerfile-base @@ -39,13 +39,9 @@ RUN git lfs install --skip-repo && \ pip3 install -U --no-cache-dir pydantic==1.10.10 RUN if [ "$TORCH_CUDA_ARCH_LIST" = "9.0+PTX" ] ; then \ - pip install "pybind11[global]"; \ - git clone https://github.com/Dao-AILab/flash-attention.git; \ - cd flash-attention; \ - git checkout v2.7.4.post1; \ - cd hopper; \ - FLASH_ATTENTION_DISABLE_SM80=TRUE MAX_JOBS=${MAX_JOBS} python setup.py install; \ - cd ../..; \ + curl -L -O https://d1dttdx32dkk5p.cloudfront.net/fa3/cu${CUDA}/torch-${PYTORCH_VERSION}/flash_attn-3.0.0b1-cp311-cp311-linux_x86_64.whl; \ + pip3 install --no-cache-dir flash_attn-3.0.0b1-cp311-cp311-linux_x86_64.whl; \ + rm flash_attn-3.0.0b1-cp311-cp311-linux_x86_64.whl; \ elif [ "$PYTORCH_VERSION" = "2.7.0" ] ; then \ pip3 install flash-attn==2.7.4.post1; \ fi