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

update other platforms' Dockerfile

Browse Source
This commit is contained in:
starkwj
2026-01-22 12:07:03 +00:00
parent 2a571d8bc8
commit 739d074b0c
14 changed files with 371 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
CMakeLists.txt
View File

@@ -77,7 +77,8 @@ message("TORCH_NPU_PATH is ${TORCH_NPU_PATH}")
if(SOC_VERSION STREQUAL "ASCEND310P3")
file(GLOB VLLM_ASCEND_SRC
${CMAKE_CURRENT_SOURCE_DIR}/csrc/*.cpp)
${CMAKE_CURRENT_SOURCE_DIR}/csrc/*.cpp
${CMAKE_CURRENT_SOURCE_DIR}/csrc/idle_offload/shm_worker.cpp)
else()
file(GLOB VLLM_ASCEND_SRC
${CMAKE_CURRENT_SOURCE_DIR}/csrc/*.cpp

6
Dockerfile.310p
View File

@@ -51,9 +51,15 @@ RUN export PIP_EXTRA_INDEX_URL=https://mirrors.huaweicloud.com/ascend/repos/pypi
source /usr/local/Ascend/nnal/atb/set_env.sh && \
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/Ascend/ascend-toolkit/latest/`uname -i`-linux/devlib && \
export SOC_VERSION=ASCEND310P3 && \
cd /vllm-workspace/vllm-ascend/csrc/idle_offload && \
make install && make clean && \
python3 -m pip install -v -e /vllm-workspace/vllm-ascend/ --extra-index https://download.pytorch.org/whl/cpu/ && \
python3 -m pip cache purge
ENV VLLM_ASCEND_ENABLE_NZ=0 \
VLLM_WORKER_MULTIPROC_METHOD=spawn \
VLLM_ASCEND_ENABLE_IDLE_OFFLOAD=1
# Install modelscope (for fast download) and ray (for multinode)
RUN python3 -m pip install modelscope 'ray>=2.47.1' 'protobuf>3.20.0' && \
python3 -m pip cache purge

6
Dockerfile.310p.openEuler
View File

@@ -49,9 +49,15 @@ RUN export PIP_EXTRA_INDEX_URL=https://mirrors.huaweicloud.com/ascend/repos/pypi
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/Ascend/ascend-toolkit/latest/`uname -i`-linux/devlib && \
export CPLUS_INCLUDE_PATH=$CPLUS_INCLUDE_PATH:/usr/include/c++/12:/usr/include/c++/12/`uname -i`-openEuler-linux && \
export SOC_VERSION=ASCEND310P3 && \
cd /vllm-workspace/vllm-ascend/csrc/idle_offload && \
make install && make clean && \
python3 -m pip install -v -e /vllm-workspace/vllm-ascend/ --extra-index https://download.pytorch.org/whl/cpu/ && \
python3 -m pip cache purge
ENV VLLM_ASCEND_ENABLE_NZ=0 \
VLLM_WORKER_MULTIPROC_METHOD=spawn \
VLLM_ASCEND_ENABLE_IDLE_OFFLOAD=1
# Install modelscope (for fast download) and ray (for multinode)
RUN python3 -m pip install modelscope 'ray>=2.47.1' 'protobuf>3.20.0' && \
python3 -m pip cache purge

10
Dockerfile.a3
View File

@@ -20,11 +20,13 @@ FROM quay.io/ascend/cann:8.3.rc2-a3-ubuntu22.04-py3.11
ARG PIP_INDEX_URL="https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple"
ARG COMPILE_CUSTOM_KERNELS=1
ARG MOONCAKE_TAG=v0.3.7.post2
ARG SOC_VERSION="ascend910_9391"
COPY . /vllm-workspace/vllm-ascend/
# Define environments
ENV DEBIAN_FRONTEND=noninteractive
ENV COMPILE_CUSTOM_KERNELS=${COMPILE_CUSTOM_KERNELS}
ENV COMPILE_CUSTOM_KERNELS=${COMPILE_CUSTOM_KERNELS} \
SOC_VERSION=$SOC_VERSION
RUN pip config set global.index-url ${PIP_INDEX_URL}
@@ -58,9 +60,15 @@ RUN export PIP_EXTRA_INDEX_URL=https://mirrors.huaweicloud.com/ascend/repos/pypi
source /usr/local/Ascend/ascend-toolkit/set_env.sh && \
source /usr/local/Ascend/nnal/atb/set_env.sh && \
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/Ascend/ascend-toolkit/latest/`uname -i`-linux/devlib && \
cd /vllm-workspace/vllm-ascend/csrc/idle_offload && \
make install && make clean && \
python3 -m pip install -v -e /vllm-workspace/vllm-ascend/ --extra-index https://download.pytorch.org/whl/cpu/ && \
python3 -m pip cache purge
ENV VLLM_ASCEND_ENABLE_NZ=0 \
VLLM_WORKER_MULTIPROC_METHOD=spawn \
VLLM_ASCEND_ENABLE_IDLE_OFFLOAD=1
# 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

10
Dockerfile.a3.openEuler
View File

@@ -20,8 +20,10 @@ FROM quay.io/ascend/cann:8.3.rc2-a3-openeuler24.03-py3.11
ARG PIP_INDEX_URL="https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple"
ARG COMPILE_CUSTOM_KERNELS=1
ARG MOONCAKE_TAG="v0.3.7.post2"
ARG SOC_VERSION="ascend910_9391"
ENV COMPILE_CUSTOM_KERNELS=${COMPILE_CUSTOM_KERNELS}
ENV COMPILE_CUSTOM_KERNELS=${COMPILE_CUSTOM_KERNELS} \
SOC_VERSION=${SOC_VERSION}
RUN pip config set global.index-url ${PIP_INDEX_URL}
@@ -61,9 +63,15 @@ RUN export PIP_EXTRA_INDEX_URL=https://mirrors.huaweicloud.com/ascend/repos/pypi
source /usr/local/Ascend/nnal/atb/set_env.sh && \
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/Ascend/ascend-toolkit/latest/`uname -i`-linux/devlib && \
export CPLUS_INCLUDE_PATH=$CPLUS_INCLUDE_PATH:/usr/include/c++/12:/usr/include/c++/12/`uname -i`-openEuler-linux && \
cd /vllm-workspace/vllm-ascend/csrc/idle_offload && \
make install && make clean && \
python3 -m pip install -v -e /vllm-workspace/vllm-ascend/ --extra-index https://download.pytorch.org/whl/cpu/ && \
python3 -m pip cache purge
ENV VLLM_ASCEND_ENABLE_NZ=0 \
VLLM_WORKER_MULTIPROC_METHOD=spawn \
VLLM_ASCEND_ENABLE_IDLE_OFFLOAD=1
# 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

6
Dockerfile.openEuler
View File

@@ -62,9 +62,15 @@ RUN export PIP_EXTRA_INDEX_URL=https://mirrors.huaweicloud.com/ascend/repos/pypi
source /usr/local/Ascend/nnal/atb/set_env.sh && \
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/Ascend/ascend-toolkit/latest/`uname -i`-linux/devlib && \
export CPLUS_INCLUDE_PATH=$CPLUS_INCLUDE_PATH:/usr/include/c++/12:/usr/include/c++/12/`uname -i`-openEuler-linux && \
cd /vllm-workspace/vllm-ascend/csrc/idle_offload && \
make install && make clean && \
python3 -m pip install -v -e /vllm-workspace/vllm-ascend/ --extra-index https://download.pytorch.org/whl/cpu/ && \
python3 -m pip cache purge
ENV VLLM_ASCEND_ENABLE_NZ=0 \
VLLM_WORKER_MULTIPROC_METHOD=spawn \
VLLM_ASCEND_ENABLE_IDLE_OFFLOAD=1
# 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

0
Dockerfile.backup → Dockerfile_backup/Dockerfile
View File

61
Dockerfile_backup/Dockerfile.310p Normal file
View File

@@ -0,0 +1,61 @@
#
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# This file is a part of the vllm-ascend project.
#
FROM quay.io/ascend/cann:8.3.rc2-310p-ubuntu22.04-py3.11
ARG PIP_INDEX_URL="https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple"
ARG COMPILE_CUSTOM_KERNELS=1
# Define environments
ENV DEBIAN_FRONTEND=noninteractive
ENV COMPILE_CUSTOM_KERNELS=${COMPILE_CUSTOM_KERNELS}
RUN apt-get update -y && \
apt-get install -y python3-pip git vim wget net-tools gcc g++ cmake libnuma-dev && \
rm -rf /var/cache/apt/* && \
rm -rf /var/lib/apt/lists/*
WORKDIR /workspace
COPY . /vllm-workspace/vllm-ascend/
RUN pip config set global.index-url ${PIP_INDEX_URL}
# Install vLLM
ARG VLLM_REPO=https://github.com/vllm-project/vllm.git
ARG VLLM_TAG=v0.11.0
RUN git clone --depth 1 $VLLM_REPO --branch $VLLM_TAG /vllm-workspace/vllm
# In x86, triton will be installed by vllm. But in Ascend, triton doesn't work correctly. we need to uninstall it.
RUN VLLM_TARGET_DEVICE="empty" python3 -m pip install -v -e /vllm-workspace/vllm/[audio] --extra-index https://download.pytorch.org/whl/cpu/ && \
python3 -m pip uninstall -y triton && \
python3 -m pip cache purge
# Install vllm-ascend
# Append `libascend_hal.so` path (devlib) to LD_LIBRARY_PATH
RUN export PIP_EXTRA_INDEX_URL=https://mirrors.huaweicloud.com/ascend/repos/pypi && \
source /usr/local/Ascend/ascend-toolkit/set_env.sh && \
source /usr/local/Ascend/nnal/atb/set_env.sh && \
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/Ascend/ascend-toolkit/latest/`uname -i`-linux/devlib && \
export SOC_VERSION=ASCEND310P3 && \
python3 -m pip install -v -e /vllm-workspace/vllm-ascend/ --extra-index https://download.pytorch.org/whl/cpu/ && \
python3 -m pip cache purge
# Install modelscope (for fast download) and ray (for multinode)
RUN python3 -m pip install modelscope 'ray>=2.47.1' 'protobuf>3.20.0' && \
python3 -m pip cache purge
CMD ["/bin/bash"]

59
Dockerfile_backup/Dockerfile.310p.openEuler Normal file
View File

@@ -0,0 +1,59 @@
#
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# This file is a part of the vllm-ascend project.
#
FROM quay.io/ascend/cann:8.3.rc2-310p-openeuler24.03-py3.11
ARG PIP_INDEX_URL="https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple"
ARG COMPILE_CUSTOM_KERNELS=1
ENV COMPILE_CUSTOM_KERNELS=${COMPILE_CUSTOM_KERNELS}
RUN yum update -y && \
yum install -y python3-pip git vim wget net-tools gcc gcc-c++ make cmake numactl-devel && \
rm -rf /var/cache/yum
RUN pip config set global.index-url ${PIP_INDEX_URL}
WORKDIR /workspace
COPY . /vllm-workspace/vllm-ascend/
# Install vLLM
ARG VLLM_REPO=https://github.com/vllm-project/vllm.git
ARG VLLM_TAG=v0.11.0
RUN git clone --depth 1 $VLLM_REPO --branch $VLLM_TAG /vllm-workspace/vllm
# In x86, triton will be installed by vllm. But in Ascend, triton doesn't work correctly. we need to uninstall it.
RUN VLLM_TARGET_DEVICE="empty" python3 -m pip install -e /vllm-workspace/vllm/[audio] --extra-index https://download.pytorch.org/whl/cpu/ && \
python3 -m pip uninstall -y triton && \
python3 -m pip cache purge
# Install vllm-ascend
RUN export PIP_EXTRA_INDEX_URL=https://mirrors.huaweicloud.com/ascend/repos/pypi && \
source /usr/local/Ascend/ascend-toolkit/set_env.sh && \
source /usr/local/Ascend/nnal/atb/set_env.sh && \
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/Ascend/ascend-toolkit/latest/`uname -i`-linux/devlib && \
export CPLUS_INCLUDE_PATH=$CPLUS_INCLUDE_PATH:/usr/include/c++/12:/usr/include/c++/12/`uname -i`-openEuler-linux && \
export SOC_VERSION=ASCEND310P3 && \
python3 -m pip install -v -e /vllm-workspace/vllm-ascend/ --extra-index https://download.pytorch.org/whl/cpu/ && \
python3 -m pip cache purge
# Install modelscope (for fast download) and ray (for multinode)
RUN python3 -m pip install modelscope 'ray>=2.47.1' 'protobuf>3.20.0' && \
python3 -m pip cache purge
CMD ["/bin/bash"]

68
Dockerfile_backup/Dockerfile.a3 Normal file
View File

@@ -0,0 +1,68 @@
#
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# This file is a part of the vllm-ascend project.
#
FROM quay.io/ascend/cann:8.3.rc2-a3-ubuntu22.04-py3.11
ARG PIP_INDEX_URL="https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple"
ARG COMPILE_CUSTOM_KERNELS=1
ARG MOONCAKE_TAG=v0.3.7.post2
COPY . /vllm-workspace/vllm-ascend/
# Define environments
ENV DEBIAN_FRONTEND=noninteractive
ENV COMPILE_CUSTOM_KERNELS=${COMPILE_CUSTOM_KERNELS}
RUN pip config set global.index-url ${PIP_INDEX_URL}
WORKDIR /workspace
# Install Mooncake dependencies
RUN apt-get update -y && \
apt-get install -y git vim wget net-tools gcc g++ cmake libnuma-dev && \
git clone --depth 1 --branch ${MOONCAKE_TAG} https://github.com/kvcache-ai/Mooncake /vllm-workspace/Mooncake && \
cp /vllm-workspace/vllm-ascend/tools/mooncake_installer.sh /vllm-workspace/Mooncake/ && \
cd /vllm-workspace/Mooncake && bash mooncake_installer.sh -y && \
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/Ascend/ascend-toolkit/latest/`uname -i`-linux/lib64 && \
mkdir -p build && cd build && cmake .. -DUSE_ASCEND_DIRECT=ON && \
make -j$(nproc) && make install && \
rm -fr /vllm-workspace/Mooncake/build && \
rm -rf /var/cache/apt/* && \
rm -rf /var/lib/apt/lists/*
# Install vLLM
ARG VLLM_REPO=https://github.com/vllm-project/vllm.git
ARG VLLM_TAG=v0.11.0
RUN git clone --depth 1 $VLLM_REPO --branch $VLLM_TAG /vllm-workspace/vllm
# In x86, triton will be installed by vllm. But in Ascend, triton doesn't work correctly. we need to uninstall it.
RUN VLLM_TARGET_DEVICE="empty" python3 -m pip install -v -e /vllm-workspace/vllm/[audio] --extra-index https://download.pytorch.org/whl/cpu/ && \
python3 -m pip uninstall -y triton && \
python3 -m pip cache purge
# Install vllm-ascend
# Append `libascend_hal.so` path (devlib) to LD_LIBRARY_PATH
RUN export PIP_EXTRA_INDEX_URL=https://mirrors.huaweicloud.com/ascend/repos/pypi && \
source /usr/local/Ascend/ascend-toolkit/set_env.sh && \
source /usr/local/Ascend/nnal/atb/set_env.sh && \
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/Ascend/ascend-toolkit/latest/`uname -i`-linux/devlib && \
python3 -m pip install -v -e /vllm-workspace/vllm-ascend/ --extra-index https://download.pytorch.org/whl/cpu/ && \
python3 -m pip cache purge
# 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
CMD ["/bin/bash"]

71
Dockerfile_backup/Dockerfile.a3.openEuler Normal file
View File

@@ -0,0 +1,71 @@
#
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# This file is a part of the vllm-ascend project.
#
FROM quay.io/ascend/cann:8.3.rc2-a3-openeuler24.03-py3.11
ARG PIP_INDEX_URL="https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple"
ARG COMPILE_CUSTOM_KERNELS=1
ARG MOONCAKE_TAG="v0.3.7.post2"
ENV COMPILE_CUSTOM_KERNELS=${COMPILE_CUSTOM_KERNELS}
RUN pip config set global.index-url ${PIP_INDEX_URL}
WORKDIR /workspace
COPY . /vllm-workspace/vllm-ascend/
SHELL ["/bin/bash", "-c"]
RUN yum update -y && \
yum install -y git vim wget net-tools gcc gcc-c++ make cmake numactl-devel && \
git clone --depth 1 --branch ${MOONCAKE_TAG} https://github.com/kvcache-ai/Mooncake /vllm-workspace/Mooncake && \
cp /vllm-workspace/vllm-ascend/tools/mooncake_installer.sh /vllm-workspace/Mooncake/ && \
ARCH=$(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:$LD_LIBRARY_PATH && \
export CPLUS_INCLUDE_PATH=$CPLUS_INCLUDE_PATH:/usr/include/c++/12:/usr/include/c++/12/${ARCH}-openEuler-linux && \
cd /vllm-workspace/Mooncake && \
bash mooncake_installer.sh -y && \
mkdir -p build && cd build && cmake .. -DUSE_ASCEND_DIRECT=ON && \
make -j$(nproc) && make install && \
rm -fr /vllm-workspace/Mooncake/build && \
rm -rf /var/cache/yum/*
# Install vLLM
ARG VLLM_REPO=https://github.com/vllm-project/vllm.git
ARG VLLM_TAG=v0.11.0
RUN git clone --depth 1 $VLLM_REPO --branch $VLLM_TAG /vllm-workspace/vllm
# In x86, triton will be installed by vllm. But in Ascend, triton doesn't work correctly. we need to uninstall it.
RUN VLLM_TARGET_DEVICE="empty" python3 -m pip install -e /vllm-workspace/vllm/[audio] --extra-index https://download.pytorch.org/whl/cpu/ && \
python3 -m pip uninstall -y triton && \
python3 -m pip cache purge
# Install vllm-ascend
RUN export PIP_EXTRA_INDEX_URL=https://mirrors.huaweicloud.com/ascend/repos/pypi && \
source /usr/local/Ascend/ascend-toolkit/set_env.sh && \
source /usr/local/Ascend/nnal/atb/set_env.sh && \
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/Ascend/ascend-toolkit/latest/`uname -i`-linux/devlib && \
export CPLUS_INCLUDE_PATH=$CPLUS_INCLUDE_PATH:/usr/include/c++/12:/usr/include/c++/12/`uname -i`-openEuler-linux && \
python3 -m pip install -v -e /vllm-workspace/vllm-ascend/ --extra-index https://download.pytorch.org/whl/cpu/ && \
python3 -m pip cache purge
# 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
CMD ["/bin/bash"]

72
Dockerfile_backup/Dockerfile.openEuler Normal file
View File

@@ -0,0 +1,72 @@
#
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# This file is a part of the vllm-ascend project.
#
FROM quay.io/ascend/cann:8.3.rc2-910b-openeuler24.03-py3.11
ARG PIP_INDEX_URL="https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple"
ARG COMPILE_CUSTOM_KERNELS=1
ARG MOONCAKE_TAG="v0.3.7.post2"
ENV COMPILE_CUSTOM_KERNELS=${COMPILE_CUSTOM_KERNELS}
RUN pip config set global.index-url ${PIP_INDEX_URL}
WORKDIR /workspace
COPY . /vllm-workspace/vllm-ascend/
SHELL ["/bin/bash", "-c"]
RUN yum update -y && \
yum install -y git vim wget net-tools gcc gcc-c++ make cmake numactl-devel && \
git clone --depth 1 --branch ${MOONCAKE_TAG} https://github.com/kvcache-ai/Mooncake /vllm-workspace/Mooncake && \
cp /vllm-workspace/vllm-ascend/tools/mooncake_installer.sh /vllm-workspace/Mooncake/ && \
ARCH=$(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:$LD_LIBRARY_PATH && \
export CPLUS_INCLUDE_PATH=$CPLUS_INCLUDE_PATH:/usr/include/c++/12:/usr/include/c++/12/${ARCH}-openEuler-linux && \
cd /vllm-workspace/Mooncake && \
bash mooncake_installer.sh -y && \
mkdir -p build && cd build && cmake .. -DUSE_ASCEND_DIRECT=ON && \
make -j$(nproc) && make install && \
rm -fr /vllm-workspace/Mooncake/build && \
rm -rf /var/cache/yum/*
# Install vLLM
ARG VLLM_REPO=https://github.com/vllm-project/vllm.git
ARG VLLM_TAG=v0.11.0
RUN git clone --depth 1 $VLLM_REPO --branch $VLLM_TAG /vllm-workspace/vllm
# In x86, triton will be installed by vllm. But in Ascend, triton doesn't work correctly. we need to uninstall it.
RUN VLLM_TARGET_DEVICE="empty" python3 -m pip install -e /vllm-workspace/vllm/[audio] --extra-index https://download.pytorch.org/whl/cpu/ && \
python3 -m pip uninstall -y triton && \
python3 -m pip cache purge
# Install vllm-ascend
RUN export PIP_EXTRA_INDEX_URL=https://mirrors.huaweicloud.com/ascend/repos/pypi && \
source /usr/local/Ascend/ascend-toolkit/set_env.sh && \
source /usr/local/Ascend/nnal/atb/set_env.sh && \
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/Ascend/ascend-toolkit/latest/`uname -i`-linux/devlib && \
export CPLUS_INCLUDE_PATH=$CPLUS_INCLUDE_PATH:/usr/include/c++/12:/usr/include/c++/12/`uname -i`-openEuler-linux && \
python3 -m pip install -v -e /vllm-workspace/vllm-ascend/ --extra-index https://download.pytorch.org/whl/cpu/ && \
python3 -m pip cache purge
# 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
CMD ["/bin/bash"]

2
csrc/idle_offload/Makefile
View File

@@ -6,7 +6,7 @@ ASCEND_HOME := /usr/local/Ascend/ascend-toolkit/latest
INCLUDES := -I$(ASCEND_HOME)/include -Iinclude
LIBS := -L$(ASCEND_HOME)/lib64 -lascendcl
CXXFLAGS := $(INCLUDES)
CXXFLAGS := $(INCLUDES) -O2
LDFLAGS := $(LIBS)
PREFIX ?= /usr/local

2
vllm_ascend/worker/worker_v1.py
View File

@@ -358,7 +358,7 @@ class NPUWorker(WorkerBase):
def try_reload_vram(self) -> tuple[bool, bool]:
allocator = CaMemAllocator.get_instance()
return allocator.try_reload_vram(tags=None)
def vnpu_unlock_gpu(self) -> None:
allocator = CaMemAllocator.get_instance()
allocator.vnpu_unlock_gpu()