Model Building and Training

Building

Programming Forms

Provide dynamic diagrams, static diagrams, dynamic and static unified programming form, so that developers can take into account the development efficiency and execution performance.

Data Loading and Processing

Two data processing models, Data Processing Pipeline and Data Processing Lightweight.

Model Building

Efficiently build and manage complex neural network architectures by utilizing the ideas of functional and object fusion programming.

Model Library

Reference cases of model implementations based on various types of kits.

Training

Training Processes

Build a complete training process that includes dataset preprocessing, model creation, defining loss functions and optimizers, training and saving the model.

Distributed Parallelism

Distributed parallelism reduces the need for hardware such as memory and computational performance, providing powerful computational capabilities and performance advantages for processing large-scale data and complex models.

Customized Higher-Order Programming

When the advanced methods provided by the framework can not meet certain scenarios, or have high performance requirements, you can use a customized method to add or modify certain processes to meet development or debugging needs.

Debugging and Optimization

Training High Availability

Failure recovery during training and other high availability functions.

Model Debugging

Model debugging methods and examples such as Dump, feature value detection, etc.

Model Tuning

Model tuning methods and cases, such as graph-kernel fusion, memory reuse, etc.