Beginner’s guide
User’s Guide
API References
开发者指南
论坛
官网

GitHub
Bilibili

How to install MegEngine
用户迁移指南
- MegEngine for NumPy users
- MegEngine for PyTorch users
Summary of common problems
- Frequently Asked Questions about Video Memory Usage
- Frequently Asked Questions about Model Reproduction

Model Development (Basics)

Deep understanding of Tensor data structure
Use Functional operations and calculations
Use Data to build the input pipeline
Use Module to define the model structure
- Module base class concept and interface introduction
Basic principles and use of Autodiff
- Autodiff 高阶使用
Use Optimizer to optimize parameters
- Parameter optimization advanced configuration
Save and Load Models (S&L)
Use Hub to publish and load pre-trained models

Model development (advanced article)

Saving memory by recomputing (Recomputation)
- Memory optimization using DTR
- 使用 Sublinear 进行显存优化
Distributed Training
Quantization
- Explanation of the principle of quantification scheme
Automatic mixing accuracy (AMP)
- 使用 NHWC 格式进一步提速
Model performance data generation and analysis (Profiler)
使用 TracedModule 发版
Just-in-time compilation (JIT)

Reasoning and deployment

Model Deployment Overview and Process Recommendations
Deploy the model with MegEngine Lite
MegEngine Lite 使用接口
- MegEngine Lite C++ 接口介绍
- MegEngine Lite python 接口介绍
使用 MegEngine Lite 部署模型进阶
使用 Load and run 测试与验证模型

Tools and plug-ins

Statistics and visualization of parameters and calculations
MegEngine 模型可视化
RuntimeOpr instructions
Custom Op

On this page

文档目录

性能优化¶

文档目录¶

输入输出内存拷贝优化
- Device IO 优化
- 输入输出拷贝优化
执行性能优化

previous

使用 MegEngine Lite 部署模型进阶

next

输入输出内存拷贝优化