What is 昇思MindSpore?
Through community collaboration, 昇思MindSpore builds an open AI architecture tailored for all scenarios, offering optimal compatibility with Ascend chips and support for multiple processor architectures. This provides algorithm engineers and data scientists with a development-friendly, highly efficient, and flexibly deployable experience, helping to drive the flourishing growth of the AI software–hardware application ecosystem.
Key features
Native distributed training: Built-in with various parallelization capabilities required for large-model training, it offers simple and easy-to-use configuration interfaces for distributed strategies, helping developers quickly achieve high-performance distributed training of large models.
Enables efficient MXFP8 low-precision training on Ascend 950 generations: Optimized for next-generation Ascend hardware, it supports low-precision training in MXFP8 format, significantly improving both training efficiency and energy efficiency.
Use cases and audience
Algorithm engineers and data scientists
- An open AI framework built for all scenarios, delivering optimal Ascend compatibility and support for multiple processor architectures, providing algorithm engineers and data scientists with a development-friendly, highly efficient, and flexibly deployable experience.
- It incorporates multiple parallel computing capabilities required for large model training and provides a simple, easy-to-use interface for configuring distributed strategies, helping developers quickly achieve high-performance distributed training of large models.
- The AI4S integrated computing framework maximizes hardware potential.
AI model code, training datasets (such as CIFAR-10), natural language text, etc.
Completed AI models, inference results (such as image classification labels or text sentiment categories)
Frequently asked questions
Does installing MindSpore require access to the public internet?
Yes, installing MindSpore requires access to the public network; if you are in an internal network environment, please ensure that your network connection is configured correctly.
Does MindSpore support distributed training for large models?
Supported. MindSpore comes with multiple parallelism capabilities required for training large models and provides a simple, easy-to-use interface for configuring distributed strategies for large models.
How to use
- 01
Visit the MindSpore official website to obtain the installation commands.
- 02
Ensure that your network connection is properly configured (installation requires access to the public internet).
- 03
Use cloud platforms such as ModelArts or the Qizhi AI Collaboration Platform to quickly create and deploy models.
- 04
Refer to the tutorial resources to practice common tasks like image classification or sentiment analysis.
Official website & sources
Sources
Updated on 2026-10-31
