MindSpore Programming Guide

Overview

• Overall Architecture
• MindSpore API Overview

Design

• Technical White Paper
• MindSpore Automatic Differentiation
• Distributed Training Design
• MindSpore IR (MindIR)
• Design of Visualization↗
• Glossary

Quickstart

• Implementing Simple Linear Function Fitting↗
• Implementing an Image Classification Application↗

Basic Concepts

• DataType
• Tensor
• Parameter
• Operators
• Cell
• Dataset

Data Pipeline

• Quick Start of Dataset
• Loading Dataset
• Processing Data
• Advanced Usage of Pipeline
• Data Iteration

Build the Network

• Constructing Single Operator Network and Multi-layer Network
• Initializer
• Network Parameters
• Using the Process Control Statement
• Parameter Passing
• Loss Function
• Gradient Operation
• Construct Constants In the Network
• Operation Overloading
• Optimization Algorithms

Model Running

• Configuring Running Information
• Running Mode
• Save and Load Models
• Application of Model

Inference

• Inference Model Overview
• Online Inference with Checkpoint
• Using Offline Model for Inference

Distributed Training

• Distributed Parallel Overview
• Distributed Parallel Advanced Features
• Distributed Parallel Usage Example

PyNative

• Debugging in PyNative Mode

Numpy

• NumPy Interfaces in MindSpore

Function Debugging

• Reading IR
• Debugging in PyNative Mode↗
• Using Dump in the Graph Mode
• Custom Debugging Information
• Incremental Operator Build

Performance Optimization

• Enabling Mixed Precision
• AutoTune
• Dataset AutoTune for Dataset Pipeline
• Enabling Offload for Dataset
• Gradient Accumulation Algorithm
• Debugging performance with Profiler↗

Advanced Features

• Second Order Optimizer
• Graph Kernel Fusion
• Quantization Aware Training

Application

• Computer Vision
• Natural Language Processing
• High Performance Computing
• Using MindSpore on the Cloud