Learning Wizard

Intended Audience

This document provides guidance for developers to use AscendCL APIs to build, modify, compile, and execute computational graphs. With this document, you will:

• Understand how to build a graph using operator prototypes or parsing an existing model.
• Understand typical scenarios of compiling and running graphs, and be familiar with AscendCL APIs used in different scenarios.

Personnel who are familiar with the basic architecture and features of CANN, capable of developing C++/C language programs, and have a basic understanding of machine learning and deep learning knowledge can better understand this document.

Development Workflow

The process is described as follows:

1. Before graph build, you need to set up the environment, install the corresponding CANN package, analyze the network structure, and develop custom operators and deploy them in the hardware environment.
2. Build a graph by using operator prototypes or directly parsing the original model.
3. Modify the graph. If you want to directly change the graph structure to the expected one, you can modify the graph based on the constructed graph for optimization.
4. Build and run the graph. You can build the modified graph into an offline model that adapts to the Ascend AI Processor, load the model through AscendCL APIs for inference. Alternatively, you can also directly run the graph to obtain the graph execution result.