Supported AI Frameworks in ModelArts Standard

For the development environment Notebook, training jobs, and model inference (model management and deployment), ModelArts Standard supports various AI frameworks and versions. Refer to the following sections.

Unified Image List

ModelArts provides a unified image for Arm+Ascend specifications, including MindSpore and PyTorch. These images are suitable for the Standard development environment, model training, and service deployment. Refer to the table below for more details.

For the URL of the images and the included dependencies, refer to ModelArts Unified Image List.

**Table 1** MindSpore
Preset Image	Supported Chip	Application Scope	Applicable Region
mindspore_2.2.0-cann_7.0.1-py_3.9-euler_2.10.7-aarch64-snt9b	Ascend snt9b	Notebook, training, and inference deployment	CN-Hong Kong

**Table 2** PyTorch
Preset Image	Supported Chip	Application Scope	Applicable Region
pytorch_2.1.0-cann_7.0.1-py_3.9-euler_2.10.7-aarch64-snt9b	Ascend snt9b	Notebook, training, and inference deployment	CN-Hong Kong
pytorch_1.11.0-cann_7.0.1-py_3.9-euler_2.10.7-aarch64-snt9b	Ascend snt9b	Notebook, training, and inference deployment	CN-Hong Kong

Development Environment Notebook

The image and versions supported by development environment notebook instances vary based on runtime environments.

**Table 3** Images supported by notebook of the new version
Image	Description	Supported Chip	Remote SSH	Online JupyterLab
pytorch1.8-cuda10.2-cudnn7-ubuntu18.04	CPU- or GPU-powered public image for general algorithm development and training, with built-in AI engine PyTorch 1.8	CPU/GPU	Yes	Yes
mindspore1.7.0-cuda10.1-py3.7-ubuntu18.04	CPU and GPU general algorithm development and training, preconfigured with AI engine MindSpore1.7.0 and cuda 10.1	CPU/GPU	Yes	Yes
mindspore1.7.0-py3.7-ubuntu18.04	CPU general algorithm development and training, preconfigured with AI engine MindSpore1.7.0	CPU	Yes	Yes
pytorch1.10-cuda10.2-cudnn7-ubuntu18.04	CPU and GPU general algorithm development and training, preconfigured with AI engine PyTorch1.10 and cuda10.2	CPU/GPU	Yes	Yes
tensorflow2.1-cuda10.1-cudnn7-ubuntu18.04	CPU- or GPU-powered public image for general algorithm development and training, with built-in AI engine TensorFlow 2.1	CPU/GPU	Yes	Yes
conda3-ubuntu18.04	Clean user customized base image only include conda	CPU	Yes	Yes
pytorch1.4-cuda10.1-cudnn7-ubuntu18.04	CPU- or GPU-powered public image for general algorithm development and training, with built-in AI engine PyTorch 1.4	CPU/GPU	Yes	Yes
tensorflow1.13-cuda10.0-cudnn7-ubuntu18.04	GPU-powered public image for general algorithm development and training, with built-in AI engine TensorFlow 1.13.1	GPU	Yes	Yes
conda3-cuda10.2-cudnn7-ubuntu18.04	Clean user customized base image include cuda10.2, conda	CPU	Yes	Yes
spark2.4.5-ubuntu18.04	CPU-powered algorithm development and training, preconfigured with PySpark 2.4.5 and can be attached to preconfigured Spark clusters including MRS and DLI	CPU	No	Yes
mindspore1.2.0-cuda10.1-cudnn7-ubuntu18.04	GPU-powered public image for algorithm development and training, with built-in AI engine MindSpore-GPU	GPU	Yes	Yes
mindspore1.2.0-openmpi2.1.1-ubuntu18.04	CPU-powered public image for algorithm development and training, with built-in AI engine MindSpore-CPU	CPU	Yes	Yes

Training Jobs

The supported AI engines and their corresponding versions for training are as follows when creating a training job.

The built-in training engines are named in the following format:

<Training engine name_version>-[cpu | <cuda_version | cann_version >]-<py_version>-<OS name_version>-< x86_64 | aarch64>

**Table 4** AI engines supported by training jobs
Runtime Environment	CPU Architecture	OS Version	AI Engine and Version	Supported CUDA or Ascend Version
TensorFlow	x86_64	Ubuntu 18.04	tensorflow_2.1.0-cuda_10.1-py_3.7-ubuntu_18.04-x86_64	cuda10.1
PyTorch	x86_64	Ubuntu 18.04	pytorch_1.8.0-cuda_10.2-py_3.7-ubuntu_18.04-x86_64	CUDA 10.2
MPI	x86_64	Ubuntu 18.04	mindspore_1.3.0-cuda_10.1-py_3.7-ubuntu_1804-x86_64	CUDA 10.1
Horovod	x86_64	Ubuntu 18.04	horovod_0.20.0-tensorflow_2.1.0-cuda_10.1-py_3.7-ubuntu_18.04-x86_64	CUDA 10.1
Horovod	x86_64	Ubuntu 18.04	horovod_0.22.1-pytorch_1.8.0-cuda_10.2-py_3.7-ubuntu_18.04-x86_64	CUDA 10.2

Supported AI engines vary depending on regions.

Supported AI Engines for Inference

If you import a preset image from a template or OBS to create a model, you can select the AI engines and versions in the table below.

Runtime environments marked as recommended is sourced from unified images, which will be the mainstream inference base image in the future. The unified image contains more comprehensive installation packages. For details, see Base Inference Images.
Images of the old version will be discontinued. Use unified images instead.
The base images to be removed are no longer maintained.
The naming convention for the unified runtime image is as follows: <AI engine name and version> - <Hardware and version: CPU or CUDA or CANN> - <Python version> - <OS version> - <CPU architecture>.

**Table 5** Supported AI engines and their runtime environments
Engine	Runtime Environment	Remarks
TensorFlow	python3.6 python2.7 (unavailable soon) tf1.13-python3.6-gpu tf1.13-python3.6-cpu tf1.13-python3.7-cpu tf1.13-python3.7-gpu tf2.1-python3.7 (unavailable soon) tensorflow_2.1.0-cuda_10.1-py_3.7-ubuntu_18.04-x86_64 (recommended)	TensorFlow 1.8.0 is used in python2.7 and python3.6. The model can run on both CPUs and GPUs when using python3.6, python2.7, or tf2.1-python3.7. For other runtime values, if the suffix contains cpu or gpu, the model can run only on CPUs or GPUs. The default runtime is python2.7.
Spark_MLlib	python2.7 (unavailable soon) python3.6 (unavailable soon)	Spark_MLlib 2.3.2 is used in python2.7 and python3.6. The default runtime is python2.7. python 2.7 and python 3.6 can only be used to run models applicable to CPU.
Scikit_Learn	python2.7 (unavailable soon) python3.6 (unavailable soon)	Scikit_Learn 0.18.1 is used in python2.7 and python3.6. The default runtime is python2.7. python 2.7 and python 3.6 can only be used to run models applicable to CPU.
XGBoost	python2.7 (unavailable soon) python3.6 (unavailable soon)	XGBoost 0.80 is used in python2.7 and python3.6. The default runtime is python2.7. python 2.7 and python 3.6 can only be used to run models applicable to CPU.
PyTorch	python2.7 (unavailable soon) python3.6 python3.7 pytorch1.4-python3.7 pytorch1.5-python3.7 (unavailable soon) pytorch_1.8.0-cuda_10.2-py_3.7-ubuntu_18.04-x86_64 (recommended)	PyTorch 1.0 is used in python2.7, python3.6, and python3.7. The model can run on both CPUs and GPUs when using python2.7, python3.6, python3.7, pytorch1.4-python3.7, or pytorch1.5-python3.7. The default runtime is python2.7.
MindSpore	aarch64 (recommended)	aarch64 can only be used to run models on Snt3 chips.