ModelArts Official Images
ModelArts offers official images for various frameworks. These images include all necessary software and dependencies. By using official images, you can quickly start notebook instances, model training, or real-time inference, reduce configuration time, and focus on model development and training, thereby improving development efficiency.
ModelArts official images comply with certain naming conventions. You can learn the basic information about an image based on its name. Typically, image names contain the fixed fields below. You are advised to follow this naming convention when creating custom images.
Taking PyTorch images as an example, the naming conventions for NPU and GPU images are described below. Naming conventions for other framework images are similar.
| Official Image Name | Supported Chip | CANN/CUDA Version | Image Name Interpretation |
|---|---|---|---|
| pytorch_2.7.1-cann_8.3.rc1-py_3.11-hce_2.0.2509-aarch64-snt9b | Ascend Snt9b | CANN 8.3.rc1 (NPU) |
|
| pytorch_1.12.1-cuda_10.2-py_3.9.11-ubuntu_20.04-x86_64 | GPU | CUDA 10.2 (GPU) |
|
Introduction to Framework Images
ModelArts provides official images based on different machine learning frameworks. The following describes the official images based on mainstream frameworks such as PyTorch, MindSpore, VeOmni, AReaL, MindSpeed-LLM, and verl.
ModelArts comes with a rich set of pre-built official PyTorch images, integrating PyTorch and its ecosystem components (Torchvision, Torchaudio). As one of the popular open-source frameworks in the deep learning field, PyTorch is widely favored for its dynamic graph mechanism, Pythonic design, and active community ecosystem. You can choose the appropriate image version according to your project requirements to quickly build a deep learning environment.
- Multi-hardware support
- Huawei Ascend NPU: integrates the torch_npu plug-in to support mixed precision training and distributed training.
- GPU: supports CUDA acceleration and provides optimization libraries such as cuDNN and NCCL.
- Pre-installed dependencies in the image:
- Common scientific computing libraries: NumPy, SciPy, and Pandas
- Visual processing libraries: OpenCV and Pillow
- Model deployment tools: ONNX Runtime and TensorBoard
- Development tools: JupyterLab, IPython, tqdm, and more
Scenario
- AI development and experiments: Quickly design, debug, and visualize model prototypes in Notebook.
- Training and tuning: A single device or multiple devices are used to perform large-scale data training in a training job.
- Deployment and servitization: You can deploy a trained model as a RESTful API through a real-time service.
- Education and training: A stable and consistent lab environment is provided to facilitate course teaching and skill training.
MindSpore is a deep learning framework in all scenarios, aiming to achieve easy development, efficient execution, and all-scenario unified deployment. MindSpore official images have MindSpore pre-installed and provide multiple combinations of Python, CANN, and EulerOS versions. They support Ascend accelerator card scenarios.
Highlights
- Unified dynamic and static graph coding: Provides a unified coding approach for both dynamic and static graphs, allowing you to switch between dynamic and static graph modes by changing just a single line of code.
- Set set_context(mode=PYNATIVE_MODE) to switch to the dynamic graph mode.
- Set set_context (mode=GRAPH_MODE) to switch to the static graph mode.
- Integrated AI and scientific computing: Extends support for flexible automatic differentiation programming capabilities, including advanced differentiation features such as forward-mode and higher-order differentiation. You can leverage this to implement integrated programming for scientific computing and AI models.
- All-scenario unified deployment: Supports unified model deployment across cloud, edge, and end devices, enabling a single codebase to run across multiple scenarios.
Scenario
- AI development and experiments: Quickly design, debug, and visualize model prototypes in Notebook.
- Training and tuning: A single device or multiple devices are used to perform large-scale data training in a training job.
- Deployment and servitization: You can deploy a trained model as a RESTful API through a real-time service.
VeOmni is an open-source full-modal distributed training framework developed by the seed team and built on PyTorch. With models as the core, the framework decouples distributed parallel logic from the model computation process, supports flexible combination of multiple parallel policies (such as FSDP, SP, and EP), and can be efficiently extended to training scenarios of ultra-long sequences and large-scale MoE models. VeOmni provides lightweight, full-modal APIs to streamline access to multimodal codecs. It incorporates system-level optimizations, including dynamic batch processing and efficient operators, to boost both training efficiency and stability.
Highlights
- Decoupling between model computation and parallelism: Based on PyTorch's native distributed capabilities, the framework decouples parallelism logic from model computation, supporting flexible combinations of Fully Sharded Data Parallel (FSDP), Sequence Parallelism (SP), and Expert Parallelism (EP).
- Full-modal access: Lightweight full-modal APIs are provided to simplify the access process of multimodal (such as visual and voice) codecs and support complex multimodal model training.
- Ascend operator-level optimization: The framework is deeply optimized for PyTorch NPU (torch_npu), integrating dynamic batching and high-performance fused operators to enhance computing throughput across full-modal scenarios.
- Ultra-large-scale scalability: The framework enables efficient training of ultra-long sequences and is optimized for large-scale MoE architectures to ensure training stability.
Scenario
- Foundation model pre-training: Supports efficient distributed training for ultra-long sequences and large-scale MoE architectures.
- Multimodal model development: Simplifies the integration and training workflows for visual, speech, and other multimodal codecs.
- Ascend NPU accelerated training: Deeply adapts to torch_npu, fully leveraging Ascend compute for multimodal model training.
As one of the mainstream open-source frameworks in the reinforcement learning field, AReaL is widely favored due to its fully asynchronous training and inference and active community ecosystem. This image combines AReaL with its ecosystem tools, vLLM and vLLM-Ascend, and supports flexible mixes of AReaL, Python, and CANN versions for Huawei Ascend NPUs. Choose the right image version based on your needs.
Highlights
- Common scientific computing libraries: NumPy and Pandas
- Inference frameworks: vLLM and vLLM-Ascend
- Model deployment tool: TensorBoard
Scenario
Reinforcement learning has three main scenarios: developing algorithms and running experiments in notebook instances, conducting large-scale training jobs, and handling end-to-end tasks that combine both training and inference.
MindSpeed-LLM is a distributed training suite for LLMs built on the Ascend computing ecosystem. It deeply integrates the MindSpeed acceleration library and Megatron-LM core architecture to provide end-to-end large model training solutions for partners. This image integrates distributed pre-training, instruction fine-tuning, and full-process toolchain (data preprocessing, weight conversion, operator acceleration, and baseline evaluation) to implement an out-of-the-box Huawei Ascend NPU development environment.
Highlights
- Ascend native acceleration: MindSpeed is deeply integrated to optimize the instruction set for torch_npu and support efficient mixed precision training and operator fusion.
- Distributed training: Based on the Megatron-LM core, multi-dimensional parallelism policies such as tensor parallelism (TP) and pipeline parallelism (PP) are supported.
- Full-stack toolchain: It provides built-in weight conversion tools (supporting format conversion of mainstream open-source models to NPU-compatible formats), distributed data preprocessing, and Ray-based resource scheduling.
- Ecosystem compatibility: Mainstream LLM architectures are supported, and the PEFT and Transformers libraries are integrated to facilitate the migration of open-source community models.
Scenario
- LLM pre-training uses the Megatron-LM core architecture to support distributed training for models with hundreds of billions of parameters.
- Instruction fine-tuning and full-pipeline training come with built-in tools for data preprocessing, weight conversion, and operator acceleration. These support complete end-to-end workflows for training foundation models.
- Open-source model migration works with foundation model architectures and allows you to convert open-source community models into NPU format for training.
verl is an easy-to-use and high-performance open-source framework in the reinforcement learning field. With its modular algorithm library and native support for distributed training, verl is widely used in reinforcement learning scenarios in scientific research and industry. This image deeply integrates the verl core framework, covering the entire process of reinforcement learning algorithm R&D, policy training, and simulation verification.
Highlights
- Modular algorithm library: Provides plug-and-play reinforcement learning algorithm modules, supporting flexible combination and extension of mainstream algorithms such as PPO and GRPO.
- Native distributed training: Built on Ray and FSDP to construct an efficient distributed training framework, supporting parallel training for large-scale Actor-Critic models.
- Training-inference integration: Integrates the vLLM inference engine to achieve seamless connection between training and inference, reducing model loading and switching overhead.
Scenario
- Reinforcement learning algorithm development and experiment: Quickly design and verify reinforcement learning algorithms in Notebook.
- Training and tuning: A single PU or multiple PUs are used to perform large-scale reinforcement learning training in a training job.
- Education and training: A stable and consistent lab environment is provided to facilitate course teaching and skill training.
Official Images
ModelArts uploads images for different frameworks to the ModelArts console or SWR. The official image list is as follows.
The images supported by ModelArts vary across regions. For details about how to view an image, see Method 1: Viewing or Using Official Images on the Console.
PyTorch
Application scope: notebook, training, and inference deployment
| Image Name | Supported Chip | CANN/CUDA Version | Image Address |
|---|---|---|---|
| pytorch_2.10.0-cann_9.1.1-py_3.11-hce_3.0.2608.1-aarch64-a5 | Ascend A5 | CANN 9.1.1 | |
| pytorch_1.12.1-cuda_10.2-py_3.9.11-ubuntu_20.04-x86_64 | GPU | CUDA 10.2 | |
| pytorch_2.7.1-cann_8.5.2-py_3.12-hce_2.0.2512-aarch64-snt9b | Ascend Snt9b | CANN 8.5.2 | |
| pytorch_2.7.1-cann_8.5.2-py_3.12-hce_2.0.2512-aarch64-snt9b23 | Ascend Snt9b2x (for example, Snt9b23) | CANN 8.5.2 | |
| pytorch_2.7.1-cann_8.3.rc1-py_3.11-hce_2.0.2509-aarch64-snt9b | Ascend Snt9b | CANN 8.3.rc1 | |
| pytorch_2.7.1-cann_8.3.rc1-py_3.11-euler2.10.11-aarch64-snt9b | Ascend Snt9b | CANN 8.3.rc1 | |
| pytorch_2.6.0-cann_8.2.rc1-py_3.11-euler_2.10.11-aarch64-snt9b | Ascend Snt9b | CANN 8.2.rc1 | |
| pytorch_2.1.0-cuda_12.1-py_3.10.6-ubuntu_22.04-x86_64 | GPU | CUDA 12.1 | |
| pytorch_2.1.0-cann_8.1.rc1-py_3.10-euler_2.10.11-aarch64-snt9b | Ascend Snt9b | CANN 8.1.rc1 | |
| pytorch_2.1.0-cann_8.0.rc3-py_3.9-euler_2.10.10-aarch64-snt9b | Ascend Snt9b | CANN 8.0.rc3 | |
| pytorch_2.1.0-cann_8.0.rc2-py_3.9-euler_2.10.7-aarch64-snt9b | Ascend Snt9b | CANN 8.0.rc2 | |
| pytorch_2.1.0-cann_8.0.0-py_3.10-euler_2.10.11-aarch64-snt9b | Ascend Snt9b | CANN 8.0.0 | |
| pytorch_1.11.0-cann_8.0.rc2-py_3.9-euler_2.10.7-aarch64-snt9b | Ascend Snt9b | CANN 8.0.rc2 | |
| pytorch_2.9.0-cann_8.5.2-py_3.11-hce_3.0.2606.1-aarch64-snt9b | Ascend Snt9b | CANN 8.5.2 | swr.<region>.myhuaweicloud.com/atelier/pytorch_ascend:pytorch_2.9.0-cann_8.5.2-py_3.11-hce_3.0.2606.1-aarch64-snt9b-20260811155239-bb77381 Example: CN-Hong Kong swr.cn-ap-southeast-1.myhuaweicloud.com/atelier/pytorch_ascend:pytorch_2.9.0-cann_8.5.2-py_3.11-hce_3.0.2606.1-aarch64-snt9b-20260811155239-bb77381 |
| pytorch_2.9.0-cann_8.5.2-py_3.11-hce_3.0.2606.1-aarch64-snt9b23 | Ascend Snt9b2x (for example, Snt9b23) | CANN 8.5.2 | swr.<region>.myhuaweicloud.com/atelier/pytorch_ascend:pytorch_2.9.0-cann_8.5.2-py_3.11-hce_3.0.2606.1-aarch64-snt9b23-20260811155239-bb77381 Example: CN-Hong Kong swr.cn-ap-southeast-1.myhuaweicloud.com/atelier/pytorch_ascend:pytorch_2.9.0-cann_8.5.2-py_3.11-hce_3.0.2606.1-aarch64-snt9b23-20260811155239-bb77381 |
| pytorch_2.7.1-cann_8.5.2-py_3.12-euler_2.10.11-aarch64-snt9b | Ascend Snt9b | CANN 8.5.2 | swr.<region>.myhuaweicloud.com/atelier/pytorch_ascend:pytorch_2.7.1-cann_8.5.2-py_3.12-euler_2.10.11-aarch64-snt9b-20260417112518-aabfd52 Example: CN-Hong Kong swr.cn-ap-southeast-1.myhuaweicloud.com/atelier/pytorch_ascend:pytorch_2.7.1-cann_8.5.2-py_3.12-euler_2.10.11-aarch64-snt9b-20260417112518-aabfd52 |
| pytorch_2.7.0-cuda_12.8-py_3.11.10-ubuntu_22.04-x86_64 | GPU | CUDA 12.8 | swr.<region>.myhuaweicloud.com/atelier/pytorch_cuda:pytorch_2.7.0-cuda_12.8-py_3.11.10-ubuntu_22.04-x86_64-20251215163925-4e5422a Example: CN-Hong Kong swr.cn-ap-southeast-1.myhuaweicloud.com/atelier/pytorch_cuda:pytorch_2.7.0-cuda_12.8-py_3.11.10-ubuntu_22.04-x86_64-20251215163925-4e5422a |
| pytorch_2.7.1-cann_8.3.rc1-py_3.11-euler_2.10.11-aarch64-snt9b | Ascend Snt9b | CANN 8.3.rc1 | swr.<region>.myhuaweicloud.com/atelier/pytorch_ascend:pytorch_2.7.1-cann_8.3.rc1-py_3.11-euler_2.10.11-aarch64-snt9b-20251208183724-e1cc47c Example: CN-Hong Kong swr.cn-ap-southeast-1.myhuaweicloud.com/atelier/pytorch_ascend:pytorch_2.7.1-cann_8.3.rc1-py_3.11-euler_2.10.11-aarch64-snt9b-20251208183724-e1cc47c |
| pytorch_2.7.1-cann_8.3.rc1-py_3.11-hce_2.0.2509-aarch64-snt9b23 | Ascend Snt9b23 | CANN 8.3.rc1 | swr.<region>.myhuaweicloud.com/atelier/pytorch_ascend:pytorch_2.7.1-cann_8.3.rc1-py_3.11-hce_2.0.2509-aarch64-snt9b23-20251205091605-e41006e Example: CN-Hong Kong swr.cn-ap-southeast-1.myhuaweicloud.com/atelier/pytorch_ascend:pytorch_2.7.1-cann_8.3.rc1-py_3.11-hce_2.0.2509-aarch64-snt9b23-20251205091605-e41006e |
| pytorch_2.7.1-cann_8.3.rc1-py_3.11-hce_2.0.2509-aarch64-snt3p | Ascend Snt3p | CANN 8.3.rc1 | swr.<region>.myhuaweicloud.com/atelier/pytorch_ascend:pytorch_2.7.1-cann_8.3.rc1-py_3.11-hce_2.0.2509-aarch64-snt3p-20251205091605-e41006e Example: CN-Hong Kong swr.cn-ap-southeast-1.myhuaweicloud.com/atelier/pytorch_ascend:pytorch_2.7.1-cann_8.3.rc1-py_3.11-hce_2.0.2509-aarch64-snt3p-20251205091605-e41006e |
| pytorch_2.6.0-cann_8.2.rc1-py_3.11-hce_2.0.2506-aarch64-snt3p | Ascend Snt3p | CANN 8.2.rc1 | swr.<region>.myhuaweicloud.com/atelier/pytorch_ascend:pytorch_2.6.0-cann_8.2.rc1-py_3.11-hce_2.0.2506-aarch64-snt3p-20250908091038-4a84562 Example: CN-Hong Kong swr.cn-ap-southeast-1.myhuaweicloud.com/atelier/pytorch_ascend:pytorch_2.6.0-cann_8.2.rc1-py_3.11-hce_2.0.2506-aarch64-snt3p-20250908091038-4a84562 |
| pytorch_2.6.0-cann_8.2.rc1-py_3.11-hce_2.0.2506-aarch64-snt9b | Ascend Snt9b | CANN 8.2.rc1 | swr.<region>.myhuaweicloud.com/atelier/pytorch_ascend:pytorch_2.6.0-cann_8.2.rc1-py_3.11-hce_2.0.2506-aarch64-snt9b-20250908091038-4a84562 Example: CN-Hong Kong swr.cn-ap-southeast-1.myhuaweicloud.com/atelier/pytorch_ascend:pytorch_2.6.0-cann_8.2.rc1-py_3.11-hce_2.0.2506-aarch64-snt9b-20250908091038-4a84562 |
| pytorch_2.6.0-cann_8.2.rc1-py_3.11-hce_2.0.2506-aarch64-snt9b23 | Ascend Snt9b23 | CANN 8.2.rc1 | swr.<region>.myhuaweicloud.com/atelier/pytorch_ascend:pytorch_2.6.0-cann_8.2.rc1-py_3.11-hce_2.0.2506-aarch64-snt9b23-20250908091038-4a84562 Example: CN-Hong Kong swr.cn-ap-southeast-1.myhuaweicloud.com/atelier/pytorch_ascend:pytorch_2.6.0-cann_8.2.rc1-py_3.11-hce_2.0.2506-aarch64-snt9b23-20250908091038-4a84562 |
MindSpore
Application scope: notebook, training, and inference deployment
| Image Name | Supported Chip | CANN/CUDA Version | Image Address |
|---|---|---|---|
| mindspore_2.0.0-cuda_11.6-py_3.9.11-ubuntu_20.04-x86_64 | GPU | CUDA 11.6 | |
| mindspore_2.7.2-cann_8.5.2-py_3.11-hce_2.0.2512-aarch64-snt9b | Ascend Snt9b | CANN 8.5.2 | |
| mindspore_2.7.2-cann_8.5.2-py_3.11-hce_2.0.2512-aarch64-snt9b23 | Ascend Snt9b2x (for example, Snt9b23) | CANN 8.5.2 | |
| mindspore_2.7.1-cann_8.3.rc1-py_3.11-euler_2.10.11-aarch64-snt9b | Ascend Snt9b | CANN 8.3.rc1 | |
| mindspore_2.7.0rc1-cann_8.2.rc1-py_3.11-euler_2.10.11-aarch64-snt9b | Ascend Snt9b | CANN 8.2.rc1 | |
| mindspore_2.6.0rc1-cann_8.1.rc1-py_3.10-euler_2.10.11-aarch64-snt9b | Ascend Snt9b | CANN 8.1.rc1 | |
| mindspore_2.4.10-cann_8.0.0-py_3.10-euler_2.10.11-aarch64-snt9b | Ascend Snt9b | CANN 8.0.0 | |
| mindspore_2.4.0-cann_8.0.rc3-py_3.9-euler_2.10.10-aarch64-snt9b | Ascend Snt9b | CANN 8.0.rc3 | |
| mindspore_2.3.0-cann_8.0.rc2-py_3.9-euler_2.10.7-aarch64-snt9b | Ascend Snt9b | CANN 8.0.rc2 | |
| mindspore_2.2.0-cann_7.0.1-py_3.9-euler_2.10.7-aarch64-snt9b | Ascend Snt9b | CANN 7.0.1 | |
| mindspore_2.7.2-cann_8.5.2-py_3.11-euler_2.10.11-aarch64-snt9b | Ascend Snt9b | CANN 8.5.2 | swr.<region>.myhuaweicloud.com/atelier/mindspore_ascend:mindspore_2.7.2-cann_8.5.2-py_3.11-euler_2.10.11-aarch64-snt9b-20260417155905-aabfd52 Example: CN-Hong Kong swr.cn-ap-southeast-1.myhuaweicloud.com/atelier/mindspore_ascend:mindspore_2.7.2-cann_8.5.2-py_3.11-euler_2.10.11-aarch64-snt9b-20260417155905-aabfd52 |
| mindspore_2.7.1-cann_8.3.rc1-py_3.11-hce_2.0.2509-aarch64-snt9b | Ascend Snt9b | CANN 8.3.rc1 | swr.<region>.myhuaweicloud.com/atelier/mindspore_ascend:mindspore_2.7.1-cann_8.3.rc1-py_3.11-hce_2.0.2509-aarch64-snt9b-20251204204157-4708bb0 Example: CN-Hong Kong swr.cn-ap-southeast-1.myhuaweicloud.com/atelier/mindspore_ascend:mindspore_2.7.1-cann_8.3.rc1-py_3.11-hce_2.0.2509-aarch64-snt9b-20251204204157-4708bb0 |
| mindspore_2.7.1-cann_8.3.rc1-py_3.11-hce_2.0.2509-aarch64-snt3p | Ascend Snt3p | CANN 8.3.rc1 | swr.<region>.myhuaweicloud.com/atelier/mindspore_ascend:mindspore_2.7.1-cann_8.3.rc1-py_3.11-hce_2.0.2509-aarch64-snt3p-20251205162951-b29ea96 Example: CN-Hong Kong swr.cn-ap-southeast-1.myhuaweicloud.com/atelier/mindspore_ascend:mindspore_2.7.1-cann_8.3.rc1-py_3.11-hce_2.0.2509-aarch64-snt3p-20251205162951-b29ea96 |
| mindspore_2.7.1-cann_8.3.rc1-py_3.11-hce_2.0.2509-aarch64-snt9b23 | Ascend Snt9b23 | CANN 8.3.rc1 | swr.<region>.myhuaweicloud.com/atelier/mindspore_ascend:mindspore_2.7.1-cann_8.3.rc1-py_3.11-hce_2.0.2509-aarch64-snt9b23-20251204204157-4708bb0 Example: CN-Hong Kong swr.cn-ap-southeast-1.myhuaweicloud.com/atelier/mindspore_ascend:mindspore_2.7.1-cann_8.3.rc1-py_3.11-hce_2.0.2509-aarch64-snt9b23-20251204204157-4708bb0 |
| mindspore_2.7.0rc1-cann_8.2.rc1-py_3.11-hce_2.0.2506-aarch64-snt3p | Ascend Snt3p | CANN 8.2.rc1 | swr.<region>.myhuaweicloud.com/atelier/mindspore_ascend:mindspore_2.7.0rc1-cann_8.2.rc1-py_3.11-hce_2.0.2506-aarch64-snt3p-20250908143531-4a84562 Example: CN-Hong Kong swr.cn-ap-southeast-1.myhuaweicloud.com/atelier/mindspore_ascend:mindspore_2.7.0rc1-cann_8.2.rc1-py_3.11-hce_2.0.2506-aarch64-snt3p-20250908143531-4a84562 |
| mindspore_2.7.0rc1-cann_8.2.rc1-py_3.11-hce_2.0.2506-aarch64-snt9b | Ascend Snt9b | CANN 8.2.rc1 | swr.<region>.myhuaweicloud.com/atelier/mindspore_ascend:mindspore_2.7.0rc1-cann_8.2.rc1-py_3.11-hce_2.0.2506-aarch64-snt9b-20250908143531-4a84562 Example: CN-Hong Kong swr.cn-ap-southeast-1.myhuaweicloud.com/atelier/mindspore_ascend:mindspore_2.7.0rc1-cann_8.2.rc1-py_3.11-hce_2.0.2506-aarch64-snt9b-20250908143531-4a84562 |
| mindspore_2.7.0rc1-cann_8.2.rc1-py_3.11-hce_2.0.2506-aarch64-snt9b23 | Ascend Snt9b23 | CANN 8.2.rc1 | swr.<region>.myhuaweicloud.com/atelier/mindspore_ascend:mindspore_2.7.0rc1-cann_8.2.rc1-py_3.11-hce_2.0.2506-aarch64-snt9b23-20250908143531-4a84562 Example: CN-Hong Kong swr.cn-ap-southeast-1.myhuaweicloud.com/atelier/mindspore_ascend:mindspore_2.7.0rc1-cann_8.2.rc1-py_3.11-hce_2.0.2506-aarch64-snt9b23-20250908143531-4a84562 |
VeOmni
Application scope: notebook and training
| Image Name | Supported Chip | CANN Version | Image Address |
|---|---|---|---|
| veomni_v0.1.6-pytorch_2.7.1-cann_8.3.rc1-py_3.11-hce_2.0.2509-aarch64-snt9b | Ascend Snt9b | CANN 8.3.rc1 |
AReaL
Application scope: notebook and training
| Image Name | Supported Chip | CANN Version | Image Address |
|---|---|---|---|
| areal_0.5.1-pytorch_2.7.1-cann_8.3.rc1-py_3.11-hce_2.0.2509-aarch64-snt9b | Ascend Snt9b | CANN 8.3.rc1 |
MindSpeed-LLM
Application scope: notebook and training
| Image Name | Supported Chip | CANN Version | Image Address |
|---|---|---|---|
| mindspeed_llm_2.2.0-pytorch_2.7.1-cann_8.2.rc2-py_3.11-hce_2.0.2509-aarch64-snt9b | Ascend Snt9b | CANN 8.2.rc2 |
verl
Application scope: notebook and training
| Image Name | Supported Chip | CANN Version | Image Address |
|---|---|---|---|
| verl_0.8.0-pytorch_2.9.0-cann_8.5.1-py_3.11-hce_2.0.2512-aarch64-snt9b | Ascend Snt9b | CANN 8.5.1 | |
| verl_0.7.0-pytorch_2.7.1-cann_8.3.rc1-py_3.11-hce_2.0.2509-aarch64-snt9b | Ascend Snt9b | CANN 8.3.rc1 |
Viewing or Using Official Images
- Method 1: Directly select an official image on the ModelArts console. The image is ready to use with zero configuration.
- Method 2: Run the docker pull command to pull an official image to the local PC. This method is suitable for local debugging and custom image creation.
Method 1: Viewing or Using Official Images on the Console
Viewing the Version List or Addresses of Official Images
- In the left navigation pane of the ModelArts console, choose Asset Management > Images (or Asset Management > Image Management).
- In the Preset Images tab (official image list), view the supported frameworks. Figure 1 Official image frameworks
- Click an official image framework card. On the Overview tab page, view the image details. On the Version List tab page, view the official image versions, including the version, image path, size, and creation time. Figure 2 Official image version list
Using an Official Image
When creating a notebook instance, training a model, or performing real-time inference on the ModelArts console, you can directly choose a preset image (official image). For details, see Creating a Notebook Instance, Creating a Training Job, or Inference Deployment (New Version).
Figure 3 Choosing an official image during notebook instance creation
Method 2: Pulling an Official Image Using docker pull
docker pull swr.cn-north-4.myhuaweicloud.com/atelier/pytorch_ascend:pytorch_2.7.1-cann_8.5.2-py_3.12-euler_2.10.11-aarch64-snt9b-20260417112518-aabfd52
For details about how to register and use an image, see Image Registration and Management.
Viewing Image Component Information
You can use either of the following methods to view details about an image component:
- Method 1: View the details in a container.
- Run the following command to start the container:
docker run -it image_name bash
Replace image_name with the actual image path. For details about how to obtain the image name, see Official Images.
- After accessing the container, run the following command to view the version information:
pip list
- Run the following command to start the container:
- Method 2: View the details in a notebook instance.
- Use an official image to create a notebook instance and access the instance in JupyterLab mode. For details, see Using JupyterLab to Develop and Debug Code Online.
- Open a terminal interface, and then run the following command to view the version information:
pip list
FAQs
Common issues and their solutions when using official images are as follows. For more information, see Image FAQs.
- How do I distinguish between a GPU image and an NPU image?
Check the keywords in the image name: Images containing cuda are GPU images. Images containing cann are NPU images.
- What should I do if docker pull displays an "unauthorized" error when pulling an image?
Ensure that you have enabled access permissions for Software Repository for Container (SWR), and that <region> in the SWR image address is replaced with the actual region (for example, replacing it with cn-southwest-2).
- What should I do if an official image cannot be found?
It might be because the image has been discontinued or is not yet supported in your region. For details about image releases, see ModelArts Container Image EOS Announcement.
- What should I do if official images do not meet my requirements?
You can create a custom image based on an official or third-party image, install the required dependencies, push it to SWR, and then use it in ModelArts. For details, see Creating a Custom Image and Using a Custom Image.
- What are the constraints on container images based on Huawei Cloud EulerOS?
The Docker version of the resource pool cluster must meet the following requirements: EulerOS 2.9 ≥ h59, EulerOS 2.10 ≥ h53, and EulerOS 2.11 ≥ h2.
If the Docker version does not meet the requirements, system compatibility issues may occur. You can log in to the resource pool cluster node to query the Docker version.
docker --version
- What are the requirements for the OS of a container image and the corresponding CCE cluster version?
The following table lists the requirements.
OS
Supported CCE Cluster Version
HCE 2.0
1.28 or later
HCE 3.0
v1.31.14-r80, v1.32.13-r50, v1.33.13-r0, v1.34.10-r0, v1.35.7-r0, v1.36.3-r0, or later
Ubuntu 22.04
1.28 or later
Ubuntu 24.04
v1.31.14-r70, v1.32.13-r40, v1.33.12-r20, v1.34.8-r20, v1.35.5-r20, v1.36.2-r10, or later
What is your overall rating for this page?
Thank you very much for your feedback. We will continue working to improve the documentation.See the reply and handling status in My Cloud VOC.
For any further questions, feel free to contact us through the chatbot.
Chatbot