TorchNPU
The plugin is a deep learning adaptation framework based on Atlas, enabling Atlas NPUs to support the PyTorch framework and providing users of the PyTorch framework with the powerful computing power of Atlas AI processors.
Quick Reference
-
TorchNPU is maintained by the https://www.hiascend.com/eng/developer/software/ai-frameworks/pytorch
-
Where to get help
- https://www.hiascend.com/en/developer/ascendhub
- https://www.hiascend.com/document/detail/en/Pytorch/latest/index/index.html
- https://www.hiascend.com/en
- Issue Feedback
Supported Tags and Dockerfile Usage
Tag Specification
Tags follow the format:
text<TorchNPU_version>-<CANN_version>-<chip>-<os>-<python_version>
| Field | value | Description |
|---|---|---|
| TorchNPU Version | 2.14.0 | For details, see the version notes in the readme. |
| CANN_version | 9.2.0 | For details, see the version notes in the readme. |
| Chip | Specific example values can be found in the CANN mirror | chip model identifier |
| OS | ubuntu22.04 / openeuler24.03 | OS distribution used for the base image |
| Python Version | py3.12 | Major Python version pre-installed in the image |
Tag list for version 26.2.0
As shown below are all the image tags of the latest 26.2.0 version released by TorchNPU. For all the historical versions, please refer to Supported Tags
2.14.0-cann9.2.0-310p-ubuntu22.04-py3.122.14.0-cann9.2.0-310p-openeuler24.03-py3.122.14.0-cann9.2.0-910b-ubuntu22.04-py3.122.14.0-cann9.2.0-910b-openeuler24.03-py3.122.14.0-cann9.2.0-a3-ubuntu22.04-py3.122.14.0-cann9.2.0-a3-openeuler24.03-py3.122.14.0-cann9.2.0-950-ubuntu22.04-py3.122.14.0-cann9.2.0-950-openeuler24.03-py3.122.13.0-cann9.2.0-310p-ubuntu22.04-py3.122.13.0-cann9.2.0-310p-openeuler24.03-py3.122.13.0-cann9.2.0-910b-ubuntu22.04-py3.122.13.0-cann9.2.0-910b-openeuler24.03-py3.122.13.0-cann9.2.0-a3-ubuntu22.04-py3.122.13.0-cann9.2.0-a3-openeuler24.03-py3.122.13.0-cann9.2.0-950-ubuntu22.04-py3.122.13.0-cann9.2.0-950-openeuler24.03-py3.122.12.0.post3-cann9.2.0-310p-ubuntu22.04-py3.122.12.0.post3-cann9.2.0-310p-openeuler24.03-py3.122.12.0.post3-cann9.2.0-910b-ubuntu22.04-py3.122.12.0.post3-cann9.2.0-910b-openeuler24.03-py3.122.12.0.post3-cann9.2.0-a3-ubuntu22.04-py3.122.12.0.post3-cann9.2.0-a3-openeuler24.03-py3.122.12.0.post3-cann9.2.0-950-ubuntu22.04-py3.122.12.0.post3-cann9.2.0-950-openeuler24.03-py3.122.11.0.post3-cann9.2.0-310p-ubuntu22.04-py3.122.11.0.post3-cann9.2.0-310p-openeuler24.03-py3.122.11.0.post3-cann9.2.0-910b-ubuntu22.04-py3.122.11.0.post3-cann9.2.0-910b-openeuler24.03-py3.122.11.0.post3-cann9.2.0-a3-ubuntu22.04-py3.122.11.0.post3-cann9.2.0-a3-openeuler24.03-py3.122.11.0.post3-cann9.2.0-950-ubuntu22.04-py3.122.11.0.post3-cann9.2.0-950-openeuler24.03-py3.122.10.0.post7-cann9.2.0-310p-ubuntu22.04-py3.122.10.0.post7-cann9.2.0-310p-openeuler24.03-py3.122.10.0.post7-cann9.2.0-910b-ubuntu22.04-py3.122.10.0.post7-cann9.2.0-910b-openeuler24.03-py3.122.10.0.post7-cann9.2.0-a3-ubuntu22.04-py3.122.10.0.post7-cann9.2.0-a3-openeuler24.03-py3.122.10.0.post7-cann9.2.0-950-ubuntu22.04-py3.122.10.0.post7-cann9.2.0-950-openeuler24.03-py3.122.9.0.post9-cann9.2.0-310p-ubuntu22.04-py3.122.9.0.post9-cann9.2.0-310p-openeuler24.03-py3.122.9.0.post9-cann9.2.0-910b-ubuntu22.04-py3.122.9.0.post9-cann9.2.0-910b-openeuler24.03-py3.122.9.0.post9-cann9.2.0-a3-ubuntu22.04-py3.122.9.0.post9-cann9.2.0-a3-openeuler24.03-py3.122.9.0.post9-cann9.2.0-950-ubuntu22.04-py3.122.9.0.post9-cann9.2.0-950-openeuler24.03-py3.122.7.1.post11-cann9.2.0-310p-ubuntu22.04-py3.122.7.1.post11-cann9.2.0-310p-openeuler24.03-py3.122.7.1.post11-cann9.2.0-910b-ubuntu22.04-py3.122.7.1.post11-cann9.2.0-910b-openeuler24.03-py3.122.7.1.post11-cann9.2.0-a3-ubuntu22.04-py3.122.7.1.post11-cann9.2.0-a3-openeuler24.03-py3.122.7.1.post11-cann9.2.0-950-ubuntu22.04-py3.122.7.1.post11-cann9.2.0-950-openeuler24.03-py3.12
Dockerfile build parameters
See dockerfile:dockerfile
| parameters | Description | Required | Reference Source | Value |
|---|---|---|---|---|
| TORCH_VERSION | Full TorchNPU version number | Yes | TorchNPU repo releases | 2.14.0 |
| CHIP_ARCH | chip architecture identifier | Yes | CANN image tag rules | Specific example values can be found in the CANN mirror |
| OS | Base image operating system | Yes | CANN image tag rules | ubuntu / openeuler |
| OS_VERSION | Operating system version | Yes | CANN image tag rules | 22.04 / 24.03 |
| PY_VERSION | Python version pre-installed in base image | Yes | CANN image tag rules | 3.11 |
| CANN_VERSION | CANN toolkit version | Yes | CANN base image repo | 9.2.0 |
| ARCH | Host hardware architecture | Yes | Environment hardware | arm / x86 |
| PY_TAG | Python package ABI tag (cp + version number) | Yes | Strictly match PY_VERSION | cp311 |
| TORCH_NPU_RELEASE_VERSION | Official TorchNPU release tag (including PyTorch version) | Yes | TorchNPU repo releases | v26.2.0-pytorch2.14.0 |
| TORCH_NPU_PATCH_TAG | TorchNPU version number in the release package name | Yes | TorchNPU repo releases | 2.14.0 |
| MANYLINUX_VER | PyPI package compatible system version | No | torch official wheel spec | manylinux_2_28 |
| PIP_MIRROR_URL | pip installation source URL (Tsinghua mirror by default) | No | PyPI mirror sources | https://pypi.tuna.tsinghua.edu.cn/simple |
Parameter Sources
-
Image tags, operating systems, and version information:https://quay.io/repository/ascend/cann?tab=tags tag
-
TORCH_NPU_RELEASE_VERSION and TORCH_NPU_PATCH_TAG parameter values are from TorchNPU Official Release Versions. Example with a wheel download URL:
[***]
- TORCH_NPU_RELEASE_VERSION is the part between
download/and/torch_npu-, e.g.v26.2.0-pytorch2.14.0. - TORCH_NPU_PATCH_TAG is the part between
torch_npu-and-cp310, e.g.2.14.0.
Quick Start
Build TorchNPU Image
Taking the construction of the 2.14.0-a3-ubuntu22.04-py3.12 image as an example:
bashdocker build \ --build-arg CHIP_ARCH=a3 \ --build-arg OS=ubuntu \ --build-arg OS_VERSION=22.04 \ --build-arg PY_VERSION=3.11 \ --build-arg CANN_VERSION=9.2.0 \ --build-arg ARCH=arm \ --build-arg PY_TAG=cp311 \ --build-arg TORCH_NPU_RELEASE_TAG=v26.2.0-pytorch2.14.0 \ --build-arg TORCH_NPU_PATCH_TAG=2.14.0 \ -t image_name:tag \ -f Dockerfile .
Note: If your build environment requires a proxy, pass proxy variables via --build-arg, for example:
bashdocker build \ --build-arg HTTP_PROXY=http://proxy.example.com:8080 \ --build-arg HTTPS_PROXY=http://proxy.example.com:8080 \ --build-arg NO_PROXY=localhost,127.0.0.1 \ ... \ -f Dockerfile .
Replace the proxy address and port with your actual environment values.
Run TorchNPU Container
bashdocker run \ --name pta_container \ --device /dev/davinci1 \ --device /dev/davinci_manager \ --device /dev/devmm_svm \ --device /dev/hisi_hdc \ -v /usr/local/dcmi:/usr/local/dcmi \ -v /usr/local/bin/npu-smi:/usr/local/bin/npu-smi \ -v /usr/local/Ascend/driver/lib64/:/usr/local/Ascend/driver/lib64/ \ -v /usr/local/Ascend/driver/version.info:/usr/local/Ascend/driver/version.info \ -v /etc/ascend_install.info:/etc/ascend_install.info \ -it ascend/pta:tag bash
Running CANN containers on 950 series aarch64 architecture products
bashdocker run \ --name pta_container \ --device /dev/davinci1 \ --device /dev/davinci_manager \ --device /dev/hisi_hdc \ --device /dev/ummu \ --device /dev/uburma \ -v /usr/local/dcmi:/usr/local/dcmi \ -v /usr/local/bin/npu-smi:/usr/local/bin/npu-smi \ -v /usr/local/Ascend/driver/lib64/:/usr/local/Ascend/driver/lib64/ \ -v /usr/local/Ascend/driver/version.info:/usr/local/Ascend/driver/version.info \ -v /etc/ascend_install.info:/etc/ascend_install.info \ -it ascend/pta:tag bash
Secondary Development
bash# Use TorchNPU image as base image and add user software FROM quay.io/ascend/torch-npu:2.14.0-cann9.2.0-910b-ubuntu22.04-py3.12 RUN apt update -y && \ apt install gcc ... ...
License/Disclaimer
See the license information for TorchNPU included in these images.
Like all container images, pre-installed software packages (Python, system libraries, etc.) may be subject to their own licenses.
The released Atlas software images are community versions; they are not intended for commercial use and serve solely as references for production practices. For detailed statements, please refer to https://www.hiascend.com/en/legal/ascendhub-download
Liability disclaimers are displayed in the image startup information and on the Atlas image platform.
镜像拉取方式
您可以使用以下命令拉取该镜像。请将 <标签> 替换为具体的标签版本。如需查看所有可用标签版本,请访问 标签列表页面。
轩辕镜像加速拉取命令点我查看更多 torch-npu 镜像标签
DockerHub 原生拉取命令
用户好评
来自真实用户的反馈,见证轩辕镜像的优质服务