Star24
88
代码
代码
Issues63
Pull Requests63
流水线
讨论
Wiki
分析
Star24
88
towncharlietowncharlie[docker]: 修改dockerfile提供方式，改为具体独立的Dockerfile
c861126e创建于 17 天前历史提交

MindIE-Motor

English | 中文

Quick Reference

MindIE-Motor

Provides one-click PD-separated deployment, flexibly adapts to multiple inference engines (vLLM, SGLang) through a cloud-native plug-in architecture, and combines high-performance scheduling with load balancing capabilities to build highly available, scalable large-scale inference services.

Tag Specification

Official pre-built image tags follow this format:

<MindIEVersion>-<ProductSeries>-<PythonVersion>-<OperatingSystem>-lts-<Architecture>
Field Example Value Description
MindIEVersion 3.0.0、 3.0.0b1、 3.0.0b2 MindIE-Motor version number
ProductSeries 800I-A2, 800I-A3 Target Atlas product series
PythonVersion py3.11 Python version
OperatingSystem Ubuntu24.04-lts, openEuler24.03-lts Base OS and distribution identifier
Architecture aarch64, x86_64 CPU architecture

3.0.0 Dockerfile Directories

Each version/hardware/OS/architecture combination has a self-contained Dockerfile

docker/mindie-motor-vllm/<tag>/Dockerfile
Tag Dockerfile
3.0.0-800I-A2-py3.11-Ubuntu24.04-lts-aarch64 Dockerfile
3.0.0-800I-A2-py3.11-Ubuntu24.04-lts-x86_64 Dockerfile
3.0.0-800I-A2-py3.11-openEuler24.03-lts-aarch64 Dockerfile
3.0.0-800I-A2-py3.11-openEuler24.03-lts-x86_64 Dockerfile
3.0.0-800I-A3-py3.11-Ubuntu24.04-lts-aarch64 Dockerfile
3.0.0-800I-A3-py3.11-Ubuntu24.04-lts-x86_64 Dockerfile
3.0.0-800I-A3-py3.11-openEuler24.03-lts-aarch64 Dockerfile
3.0.0-800I-A3-py3.11-openEuler24.03-lts-x86_64 Dockerfile

Each Dockerfile embeds the matching vllm-ascend base image (v0.18.0 series), target platform, output image tag, entrypoint script, and license agreement. No helper script, external docker files, or environment-variable selection is required.

Quick Start

Prerequisites (Optional)

Install Drivers

Build MindIE-Motor Image

Each Dockerfile clones the pinned branch and commit during the build. No local source tree or build context is required. Replace <tag> with the desired combination:

TAG="3.0.0-800I-A2-py3.11-Ubuntu24.04-lts-aarch64"

docker build --network=host \
    --platform=linux/arm64 \
    -t "mindie-motor-vllm:${TAG}" \
    -f "docker/mindie-motor-vllm/${TAG}/Dockerfile" \
    .

Each Dockerfile header comment contains the exact --platform value, source repository info, and full docker build command.

3.0.0 source mapping:

Image Version Repository Branch Commit
3.0.0 https://gitcode.com/Ascend/MindIE-PyMotor.git v3.0.0 383d1787ed3fc27aaad2db9cc5506d40c258c279

The build process:

  1. Pulls the matching vllm-ascend base image.
  2. Clones the pinned MindIE-PyMotor source into /opt/MindIE-PyMotor.
  3. Installs dependencies, compiles, and installs the motor wheel.
  4. Builds and installs the ccae_reporter observability component.
  5. Generates the entrypoint script and license agreement inline within the Dockerfile.

Run MindIE-Motor Container

Confirm that Atlas drivers are installed on the host and /dev/davinci* device nodes are available.

Minimal Verification

IMAGE_NAME="mindie-motor-vllm:3.0.0-800I-A2-py3.11-Ubuntu24.04-lts-aarch64"

docker run --rm -it \
  --device=/dev/davinci_manager \
  --device=/dev/devmm_svm \
  --device=/dev/hisi_hdc \
  --device=/dev/davinci0 \
  -v /usr/local/Ascend/driver:/usr/local/Ascend/driver:ro \
  -v /usr/local/Ascend/add-ons/:/usr/local/Ascend/add-ons/:ro \
  -v /usr/local/sbin/npu-smi:/usr/local/sbin/npu-smi:ro \
  -v /var/log/npu/:/usr/slog \
  "${IMAGE_NAME}" \
  bash -c "npu-smi info && python -c 'import motor; print(\"motor ok\")'"

Start Inference Service

Prepare boot.sh, user_config.json, and other config files before deployment. See the docker-only single-container deployment guide for the full workflow.

CONFIGMAP_PATH="/path/to/configmap"
IMAGE_NAME="mindie-motor-vllm:3.0.0-800I-A2-py3.11-Ubuntu24.04-lts-aarch64"

docker run -u root --rm --name mindie-motor \
  -e ASCEND_RUNTIME_OPTIONS=NODRV \
  -e CONFIGMAP_PATH="${CONFIGMAP_PATH}" \
  -e CONFIG_PATH=/usr/local/Ascend/pyMotor/conf \
  -e ROLE=SINGLE_CONTAINER \
  --device=/dev/davinci_manager \
  --device=/dev/devmm_svm \
  --device=/dev/hisi_hdc \
  --device=/dev/davinci0 \
  --device=/dev/davinci1 \
  -p 1025:1025 \
  -p 1026:1026 \
  -v /usr/local/Ascend/driver:/usr/local/Ascend/driver \
  -v /usr/local/Ascend/add-ons/:/usr/local/Ascend/add-ons/ \
  -v /usr/local/sbin/npu-smi:/usr/local/sbin/npu-smi \
  -v /usr/local/sbin:/usr/local/sbin \
  -v /var/log/npu/:/usr/slog \
  -v /mnt:/mnt \
  -v "${CONFIGMAP_PATH}:${CONFIGMAP_PATH}" \
  "${IMAGE_NAME}" \
  bash -c 'export POD_IP=$(grep $(hostname) /etc/hosts | cut -f1) && source ${CONFIGMAP_PATH}/boot.sh'

Common parameters:

Parameter / Env Var Description
--device=/dev/davinci{N} Map NPU devices; add davinci0, davinci1, etc. as needed
--device=/dev/davinci_manager etc. Atlas management devices, usually required
-v /usr/local/Ascend/driver:... Mount host Atlas driver directory
-v ${CONFIGMAP_PATH}:... Mount startup scripts and config directory
-p <host>:<container> Expose API ports per user_config.json
ASCEND_RUNTIME_OPTIONS=NODRV Reuse host drivers without reinstalling inside the container
CONFIGMAP_PATH Path to startup scripts inside the container
CONFIG_PATH Motor config directory, default /usr/local/Ascend/pyMotor/conf
ROLE Deployment role; use SINGLE_CONTAINER for single-container PD separation

Extend the Image

FROM mindie-motor-vllm:3.0.0-800I-A2-py3.11-Ubuntu24.04-lts-aarch64

RUN apt update -y && \
    apt install gcc ...

Supported Hardware

Chip Series Product Examples Architecture
Atlas 910 Atlas 800T A2, Atlas 900 A2 PoD ARM64 / x86_64
Atlas A3 Atlas 800T A3 ARM64 / x86_64

Image Version Notes

Version Description Notes
3.0.0 MindIE 3.0.0 Release 2026/5/6: initial release

License

See the Motor license for license information included in these images.

As with all container images, pre-installed software packages (Python, system libraries, etc.) may be subject to their own licenses.