x
This commit is contained in:
+2
-5
@@ -64,11 +64,8 @@ RUN python -m pip install \
|
|||||||
WORKDIR /opt
|
WORKDIR /opt
|
||||||
ENV FLASH_ATTENTION_TRITON_AMD_ENABLE="TRUE"
|
ENV FLASH_ATTENTION_TRITON_AMD_ENABLE="TRUE"
|
||||||
|
|
||||||
RUN git clone https://github.com/ROCm/flash-attention.git &&\
|
RUN pip install flash-attn --no-build-isolation \
|
||||||
cd flash-attention &&\
|
--index-url https://pypi.tuna.tsinghua.edu.cn/simple
|
||||||
git checkout main_perf &&\
|
|
||||||
python setup.py install && \
|
|
||||||
cd /opt && rm -rf /opt/flash-attention
|
|
||||||
|
|
||||||
# 6. Clone vLLM
|
# 6. Clone vLLM
|
||||||
RUN git clone https://github.com/vllm-project/vllm.git /opt/vllm
|
RUN git clone https://github.com/vllm-project/vllm.git /opt/vllm
|
||||||
|
|||||||
Reference in New Issue
Block a user