explicitly pin flash attention 1 to v1.0.9
Browse files- docker/Dockerfile-base +2 -1
docker/Dockerfile-base
CHANGED
@@ -38,8 +38,9 @@ WORKDIR /workspace
|
|
38 |
|
39 |
ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6+PTX"
|
40 |
|
41 |
-
RUN git clone https://github.com/
|
42 |
cd flash-attention && \
|
|
|
43 |
python3 setup.py bdist_wheel && \
|
44 |
cd csrc/fused_dense_lib && \
|
45 |
python3 setup.py bdist_wheel && \
|
|
|
38 |
|
39 |
ARG TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6+PTX"
|
40 |
|
41 |
+
RUN git clone https://github.com/Dao-AILab/flash-attention.git && \
|
42 |
cd flash-attention && \
|
43 |
+
git checkout v1.0.9 && \
|
44 |
python3 setup.py bdist_wheel && \
|
45 |
cd csrc/fused_dense_lib && \
|
46 |
python3 setup.py bdist_wheel && \
|