EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[CI] Refactor to speedup image building and CI Installation (#6708)

Browse Source 
### What this PR does / why we need it?
1. Refactor  image workflow using cache-from to speedup builds

![build](https://github.com/user-attachments/assets/02135c12-0069-44f8-a3ec-5c2b4282448a)

Simultaneously refactored all Dockerfiles by placing layers that rarely
change before those that change frequently, improving build cache hit
rate.

2. Refactor E2E test using vllm-ascend container images, to skip C
compile while no C code are changed

![e2e](https://github.com/user-attachments/assets/49f5b166-0df3-41e1-8f71-b3bbbed17cfd)

In this case, the job will only replace the source code of vllm-ascend
and install `requirements-dev.txt`, saving about 10min before tests

### Does this PR introduce _any_ user-facing change?

### How was this patch tested?

- vLLM version: v0.15.0
- vLLM main:
9562912cea

Signed-off-by: wjunLu <wjunlu217@gmail.com>
This commit is contained in:
wjunLu
2026-02-28 09:06:00 +08:00
committed by GitHub
parent 5666ce03f5
commit 84b00695f8
13 changed files with 456 additions and 213 deletions
Download Patch File Download Diff File Expand all files Collapse all files

50
Dockerfile
View File

@@ -19,46 +19,49 @@ FROM quay.io/ascend/cann:8.5.0-910b-ubuntu22.04-py3.11
ARG PIP_INDEX_URL="https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple"
ARG MOONCAKE_TAG="v0.3.8.post1"
ARG SOC_VERSION="ascend910b1"
# Define environments
ENV DEBIAN_FRONTEND=noninteractive
ENV SOC_VERSION=$SOC_VERSION \
TASK_QUEUE_ENABLE=1 \
OMP_NUM_THREADS=1
WORKDIR /workspace
COPY . /vllm-workspace/vllm-ascend/
COPY ./tools/mooncake_installer.sh /vllm-workspace/
# Install Mooncake dependencies
# Install clang-15 (for triton-ascend) and Mooncake
RUN apt-get update -y && \
apt-get install -y git vim wget net-tools gcc g++ cmake libnuma-dev libjemalloc2 && \
apt-get install -y git vim wget net-tools gcc g++ cmake libnuma-dev libjemalloc2 clang-15 && \
update-alternatives --install /usr/bin/clang clang /usr/bin/clang-15 20 && \
update-alternatives --install /usr/bin/clang++ clang++ /usr/bin/clang++-15 20 && \
git clone --depth 1 --branch ${MOONCAKE_TAG} https://github.com/kvcache-ai/Mooncake /vllm-workspace/Mooncake && \
cp /vllm-workspace/vllm-ascend/tools/mooncake_installer.sh /vllm-workspace/Mooncake/ && \
mv /vllm-workspace/mooncake_installer.sh /vllm-workspace/Mooncake/ && \
cd /vllm-workspace/Mooncake && bash mooncake_installer.sh -y && \
ARCH=$(uname -m) && \
source /usr/local/Ascend/ascend-toolkit/set_env.sh && \
export LD_LIBRARY_PATH=/usr/local/Ascend/ascend-toolkit/latest/${ARCH}-linux/devlib:/usr/local/Ascend/ascend-toolkit/latest/${ARCH}-linux/lib64:$LD_LIBRARY_PATH && \
mkdir -p build && cd build && cmake .. -DUSE_ASCEND_DIRECT=ON && \
make -j$(nproc) && make install && \
rm -fr /vllm-workspace/Mooncake/build && \
rm -rf /vllm-workspace/Mooncake/build && \
rm -rf /var/cache/apt/* && \
rm -rf /var/lib/apt/lists/*
RUN pip config set global.index-url ${PIP_INDEX_URL}
# Install modelscope (for fast download) and ray (for multinode)
RUN pip config set global.index-url ${PIP_INDEX_URL} && \
python3 -m pip install modelscope 'ray>=2.47.1,<=2.48.0' 'protobuf>3.20.0' && \
python3 -m pip cache purge
# Install vLLM
ARG VLLM_REPO=https://github.com/vllm-project/vllm.git
ARG VLLM_TAG=v0.16.0
RUN git clone --depth 1 $VLLM_REPO --branch $VLLM_TAG /vllm-workspace/vllm
# In x86, triton will be installed by vllm. But in Ascend, triton doesn't work correctly. we need to uninstall it.
RUN VLLM_TARGET_DEVICE="empty" python3 -m pip install -v -e /vllm-workspace/vllm/[audio] --extra-index https://download.pytorch.org/whl/cpu/ && \
RUN git clone --depth 1 $VLLM_REPO --branch $VLLM_TAG /vllm-workspace/vllm && \
VLLM_TARGET_DEVICE="empty" python3 -m pip install -v -e /vllm-workspace/vllm/[audio] --extra-index https://download.pytorch.org/whl/cpu/ && \
python3 -m pip uninstall -y triton && \
python3 -m pip cache purge
# Install vllm-ascend
# Append `libascend_hal.so` path (devlib) to LD_LIBRARY_PATH
ARG SOC_VERSION="ascend910b1"
ENV DEBIAN_FRONTEND=noninteractive
ENV SOC_VERSION=$SOC_VERSION \
TASK_QUEUE_ENABLE=1 \
OMP_NUM_THREADS=1
COPY . /vllm-workspace/vllm-ascend/
RUN export PIP_EXTRA_INDEX_URL=https://mirrors.huaweicloud.com/ascend/repos/pypi && \
source /usr/local/Ascend/ascend-toolkit/set_env.sh && \
source /usr/local/Ascend/nnal/atb/set_env.sh && \
@@ -66,18 +69,7 @@ RUN export PIP_EXTRA_INDEX_URL=https://mirrors.huaweicloud.com/ascend/repos/pypi
python3 -m pip install -v -e /vllm-workspace/vllm-ascend/ --extra-index https://download.pytorch.org/whl/cpu/ && \
python3 -m pip cache purge
# Install clang-15 (for triton-ascend)
RUN apt-get update -y && \
apt-get -y install clang-15 && \
update-alternatives --install /usr/bin/clang clang /usr/bin/clang-15 20 && \
update-alternatives --install /usr/bin/clang++ clang++ /usr/bin/clang++-15 20 && \
rm -rf /var/cache/apt/* && \
rm -rf /var/lib/apt/lists/*
# Install modelscope (for fast download) and ray (for multinode)
RUN python3 -m pip install modelscope 'ray>=2.47.1,<=2.48.0' 'protobuf>3.20.0' && \
python3 -m pip cache purge
# Append `libascend_hal.so` path (devlib) to LD_LIBRARY_PATH
RUN echo "export LD_PRELOAD=/usr/lib/$(uname -m)-linux-gnu/libjemalloc.so.2:$LD_PRELOAD" >> ~/.bashrc
CMD ["/bin/bash"]