Quick Start

Overview
Quick Start
Tensor
Data Loading and Processing
Building a Network
Automatic Differentiation
Model Training
Saving and Loading the Model
Accelerating with Static Graphs
Automatic Mix Precision

Data Processing

Data Sampling
Converting Dataset to MindRecord
Lightweight Data Processing
Supporting Python Objects in Dataset Pipeline
Auto Augmentation
Single-Node Data Cache
Optimizing the Data Processing

Compilation

Introduction to Graph Mode Programming
Graph Mode Syntax - Operators
Graph Mode Syntax - Python Statements
Graph Mode Syntax - Python Built-in Functions
Graph Mode - Programming Techniques

Parallel

Distributed Parallelism Overview
Distributed Parallel Startup Methods
Data Parallel
Operator-level Parallelism
Optimizer Parallel
Pipeline Parallel
Optimization Techniques
Distributed High-Level Configuration Case
- Multi-dimensional Hybrid Parallel Case Based on Double Recursive Search
- Performing Distributed Training on K8S Clusters

Debugging and Tuning

Dynamic Graph Debugging
Using Dump in the Graph Mode
Feature Value Detection
Ascend Performance Tuning
Error Reporting Analysis
DryRun

Custom programming

Custom Operators
Custom Fusion
Hook Programming

Infer

MindSpore Large Language Model Inference
Obtaining and Preparing Large Language Model Weights
Building a Large Language Model Inference Network from Scratch
Building a Parallel Large Language Model Network
Multi-device Model Weight Sharding
Model Quantization
Lite Inference Overview

High Availability

Fault Recovery
Training Process Exit Gracefully

Orange Pi

Orange Pi Development
Environment Setup Guide
Model Online Inference
Quick Start

Model Cases

Model Migration
Computer Vision
Natural Language Processing
Generative

MindSpore

»
Distributed High-Level Configuration Case
View page source

Distributed High-Level Configuration Case

Multi-dimensional Hybrid Parallel Case Based on Double Recursive Search
Performing Distributed Training on K8S Clusters

Previous Next

© Copyright MindSpore.

Built with Sphinx using a theme provided by Read the Docs.