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

fix multiproc executor determine kv cache memory & update Dockerfile

Browse Source
This commit is contained in:
starkwj
2026-04-24 08:31:54 +00:00
parent e4d898b245
commit e17006077a
10 changed files with 145 additions and 153 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
Dockerfile.310p.openEuler
View File

@@ -20,10 +20,6 @@ FROM quay.io/ascend/cann:8.5.1-310p-openeuler24.03-py3.11
ARG PIP_INDEX_URL="https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple"
ARG SOC_VERSION="ascend310p1"
ENV SOC_VERSION=$SOC_VERSION \
TASK_QUEUE_ENABLE=1 \
OMP_NUM_THREADS=1
RUN yum update -y && \
yum install -y python3-pip git vim wget net-tools gcc gcc-c++ make cmake numactl numactl-devel jemalloc && \
rm -rf /var/cache/yum
@@ -44,6 +40,11 @@ RUN VLLM_TARGET_DEVICE="empty" python3 -m pip install -e /vllm-workspace/vllm/[a
python3 -m pip cache purge
# Install vllm-ascend
ENV SOC_VERSION=$SOC_VERSION \
TASK_QUEUE_ENABLE=1 \
OMP_NUM_THREADS=1 \
VLLM_WORKER_MULTIPROC_METHOD=spawn \
VLLM_ASCEND_ENABLE_VNPU=1
# Installing vllm-ascend on x86 can pull upstream triton back in alongside triton-ascend. Remove it immediately after this step.
RUN export PIP_EXTRA_INDEX_URL=https://mirrors.huaweicloud.com/ascend/repos/pypi && \
source /usr/local/Ascend/ascend-toolkit/set_env.sh && \
@@ -66,9 +67,6 @@ RUN export PIP_EXTRA_INDEX_URL=https://mirrors.huaweicloud.com/ascend/repos/pypi
python3 -m pip install "https://vllm-ascend.obs.cn-north-4.myhuaweicloud.com/vllm-ascend/${PTA_WHEEL}" && \
python3 -m pip cache purge
ENV VLLM_WORKER_MULTIPROC_METHOD=spawn \
VLLM_ASCEND_ENABLE_VNPU=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