mindspore 2.8.0

MindSpore is a new open source deep learning training/inference framework that could be used for mobile, edge and cloud scenarios.

查看中文

• What Is MindSpore
  • Automatic Differentiation
  • Automatic Parallel
• Installation
  • Pip mode method installation
  • Source code compilation installation
  • Docker Image
• Quickstart
• Docs
• Community
  • Governance
  • Communication
• Contributing
• Maintenance phases
• Maintenance status
• Release Notes
• License

What Is MindSpore

MindSpore is a new open source deep learning training/inference framework that could be used for mobile, edge and cloud scenarios. MindSpore is designed to provide development experience with friendly design and efficient execution for the data scientists and algorithmic engineers, native support for Ascend AI processor, and software hardware co-optimization. At the meantime MindSpore as a global AI open source community, aims to further advance the development and enrichment of the AI software/hardware application ecosystem.

For more details please check out our Architecture Guide.

Automatic Differentiation

Currently, there are two automatic differentiation techniques in mainstream deep learning frameworks:

• Operator Overloading (OO): Overloading the basic operators of the programming language to encapsulate their gradient rules. Record the operation trajectory of the network during forward execution in an operator overloaded manner, then apply the chain rule to the dynamically generated data flow graph to implement automatic differentiation.
• Source Transformation (ST): This technology is evolving from the functional programming framework and performs automatic differential transformation on the intermediate expression (the expression form of the program during the compilation process) in the form of just-in-time compilation (JIT), supporting complex control flow scenarios, higher-order functions and closures.

PyTorch used OO. Compared to ST, OO generates gradient graph in runtime, so it does not need to take function call and control flow into consideration, which makes it easier to develop. However, OO can not perform gradient graph optimization in compilation time and the control flow has to be unfolded in runtime, so it is difficult to achieve extreme optimization in performance.

MindSpore implemented automatic differentiation based on ST. On the one hand, it supports automatic differentiation of automatic control flow, so it is quite convenient to build models like PyTorch. On the other hand, MindSpore can perform static compilation optimization on neural networks to achieve great performance.

The implementation of MindSpore automatic differentiation can be understood as the symbolic differentiation of the program itself. Because MindSpore IR is a functional intermediate expression, it has an intuitive correspondence with the composite function in basic algebra. The derivation formula of the composite function composed of arbitrary basic functions can be derived. Each primitive operation in MindSpore IR can correspond to the basic functions in basic algebra, which can build more complex flow control.

Automatic Parallel

The goal of MindSpore automatic parallel is to build a training method that combines data parallelism, model parallelism, and hybrid parallelism. It can automatically select a least cost model splitting strategy to achieve automatic distributed parallel training.

At present, MindSpore uses a fine-grained parallel strategy of splitting operators, that is, each operator in the figure is split into a cluster to complete parallel operations. The splitting strategy during this period may be very complicated, but as a developer advocating Pythonic, you don't need to care about the underlying implementation, as long as the top-level API compute is efficient.

Installation

Pip mode method installation

MindSpore offers build options across multiple backends:

Hardware Platform	Operating System	Status
Ascend	Linux-x86	✔️
	Linux-aarch64	✔️
GPU CUDA 11.6	Linux-x86	✔️
CPU	Linux-x86	✔️
	Linux-aarch64	✔️
	Windows-x86	✔️
	MacOS-x86	✔️
	MacOS-aarch64	✔️

For installation using pip, take CPU and Linux-x86 build version as an example:

1. Download whl from MindSpore download page, and install the package.

   pip install mindspore==2.7.1 -i https://repo.mindspore.cn/pypi/simple --trusted-host repo.mindspore.cn --extra-index-url https://repo.huaweicloud.com/repository/pypi/simple
   

2. Run the following command to verify the install.

   python -c "import mindspore;mindspore.set_device(device_target='CPU');mindspore.run_check()"
   

Use pip mode method to install MindSpore in different environments. Refer to the following documents.

• Using pip mode method to install MindSpore in Ascend environment
• Using pip mode method to install MindSpore in GPU environment
• Using pip mode method to install MindSpore in CPU environment

Source code compilation installation

Use the source code compilation method to install MindSpore in different environments. Refer to the following documents.

• Using the source code compilation method to install MindSpore in Ascend environment
• Using the source code compilation method to install MindSpore in GPU environment
• Using the source code compilation method to install MindSpore in CPU environment

Docker Image

MindSpore docker image is hosted on Huawei SWR. Use Docker to install MindSpore in different environments, refer to the following documents.

• Installing MindSpore in Ascend by Docker

Quickstart

See the Quick Start to implement the image classification.

Docs

More details about installation guide, tutorials and APIs, please see the User Documentation.

Community

Governance

Check out how MindSpore Open Governance works.

Communication

• IRC channel at #mindspore (only for meeting minutes logging purpose)
• Video Conferencing: TBD
• Mailing-list: https://mailweb.mindspore.cn/postorius/lists

Contributing

Welcome contributions. See our Contributor Wiki for more details.

Maintenance phases

Project stable branches will be in one of the following states:

State	Time frame	Summary
Planning	1 - 3 months	Features are under planning.
Development	3 months	Features are under development.
Maintained	6 - 12 months	All bugfixes are appropriate. Releases produced.
Unmaintained	0 - 3 months	All bugfixes are appropriate. No Maintainers and No Releases produced.
End Of Life (EOL)	N/A	Version no longer accepting changes.

Maintenance status

Version	Status	Initial Release Date	Next Phase	EOL Date
r2.7	Maintained	2025-08-08	Unmaintained 2026-08-08 estimated	2026-08-08
r2.6	Maintained	2025-05-19	Unmaintained 2026-05-19 estimated	2026-05-19
r2.5	Maintained	2025-02-08	Unmaintained 2026-02-08 estimated	2026-02-08
r2.4	End Of Life	2024-10-30		2025-10-30
r2.3	End Of Life	2024-07-15		2025-07-15
r2.2	End Of Life	2023-10-18		2024-10-18
r2.1	End Of Life	2023-07-29		2024-07-29
r2.0	End Of Life	2023-06-15		2024-06-15
r1.10	End Of Life	2023-02-02		2024-02-02
r1.9	End Of Life	2022-10-26		2023-10-26
r1.8	End Of Life	2022-07-29		2023-07-29
r1.7	End Of Life	2022-04-29		2023-04-29
r1.6	End Of Life	2022-01-29		2023-01-29
r1.5	End Of Life	2021-10-15		2022-10-15
r1.4	End Of Life	2021-08-15		2022-08-15
r1.3	End Of Life	2021-07-15		2022-07-15
r1.2	End Of Life	2021-04-15		2022-04-29
r1.1	End Of Life	2020-12-31		2021-09-30
r1.0	End Of Life	2020-09-24		2021-07-30
r0.7	End Of Life	2020-08-31		2021-02-28
r0.6	End Of Life	2020-07-31		2020-12-30
r0.5	End Of Life	2020-06-30		2021-06-30
r0.3	End Of Life	2020-05-31		2020-09-30
r0.2	End Of Life	2020-04-30		2020-08-31
r0.1	End Of Life	2020-03-28		2020-06-30

Release Notes

The release notes, see our RELEASE.

License

Apache License 2.0