MindStudio
This Huawei-designed full-pipeline development toolchain equips Ascend AI developers with an end-to-end solution for building Ascend AI applications, accelerating the processes of training, inference, and operator development.
- What Is MindStudio
Describes the architecture and application scenarios of MindStudio.
- Release Notes
Describes the version mapping, new features, and resolved issues of MindStudio.
- Quick Start
Provides a quick start for using the tool in the training, inference, and operator development scenarios.
Visualization Tool
Operator Development Tools
Design
- Operator Design mskpp
Allows users to enter operator expressions to predict the performance upper limit of operators in the algorithm implementation.
Development
- Operator Project msOpGen
Provides the template project generation capability, simplifies operator project setup, and assists in operator test and verification. msOpST is used to test the input and output of operators in the real hardware environment to verify the operator functions.
Debugging
- Operator Debugging msDebug
Provides native environment debugging on Ascend processors, featuring flexible variable inspection and step-by-step execution.
- Anomaly Detection msSanitizer
Provides memory, contention, uninitialization, and synchronization check capabilities and accurately locates memory issues in multi-core programs.
Tuning
- Operator Tuning msOpProf
Collects profile data from both hardware and simulation, visualized through MindStudio Insight to identify performance bottlenecks.
Run
Training Development Tools
- Model Optimization msProf
Collects CANN and NPU profile data to improve the tuning efficiency.
- Ascend PyTorch Profiler
Collects profile data in PyTorch training and online inference scenarios and outputs visualized profile data files, improving performance analysis efficiency.
- MindSpore Profiler
Collects profile data in MindSpore training and online inference scenarios and outputs visualized profile data files, improving performance analysis efficiency.
- Performance Analysis msprof-analyze
Analyzes the collected profile data and quickly identifies performance bottlenecks of Ascend devices.
- Tuning Tool Interface Library msPTI
Builds tools for NPU-based applications to analyze the performance of applications.
- Memory Analysis msMemScope
Provides multi-dimensional memory data collection, automatic diagnosis, and optimization analysis across the entire network.
Run
- Online Monitoring msMonitor
Supports both drive dump and online profile data collection, with performance monitoring and fault location capabilities in cluster scenarios.
Extension Tools
Inference Development Tools
- Model Optimization msProf
Collects CANN and NPU profile data to improve the tuning efficiency.
- Performance Analysis msprof-analyze
Analyzes the collected profile data and quickly identifies performance bottlenecks of Ascend devices.
- Serving Tuning msServiceProfiler
Supports request scheduling and model execution visualization, improving service performance analysis efficiency.
- Tuning Tool Interface Library msPTI
Builds tools for NPU-based applications to analyze the performance of applications.
- Memory Analysis msMemScope
Provides multi-dimensional memory data collection, automatic diagnosis, and optimization analysis across the entire network.
Run
Tool Extension SDK
Cases
- General Performance Issue Troubleshooting Guide
- Foundation Model Inference Accuracy Debugging Guide
- Foundation Model Training Accuracy Debugging Guide
- Case Study on Locating Foundation Model Training Performance Bottlenecks
- Debugging and Tuning Guide for Foundation Model Inference Quantization
- Porting, Debugging, and Tuning Guide for Traditional Model Inference