Learning Wizard

This document provides guidance for developers to use the Ascend Tensor Compiler (ATC) tool to convert a model into an offline model adapted to the Ascend AI Processor. With this document, you will be able to:

Convert original models from popular frameworks into offline models adapted to the Ascend AI Processor.
Customize offline models for different requirements based on the given parameters.

Familiarity with basic Linux commands and a basic understanding of machine learning and deep learning will help you use this document better.

For Beginners

ATC Introduction	Getting Started	Fundamental Features
Describes the tool architecture, workflow, and key terms.	Describes how to convert your model with ATC by taking supported models from different frameworks as examples.	Describes the basic functions of ATC, such as converting a model to a JSON file to show parameter information, adding dynamic batch size and image size support to an offline model, and customizing an offline model with various ATC options.

For Experts

AIPP	Generating a Single-Operator Model	Special Topics
Describes the definition and usage of AIPP with typical samples, including creating an AIPP configuration file based on the given template and generating required images using color space conversion (CSC).	Describes the single-operator description file, including how to generate a single-operator description file and how to convert such a file into an offline model adapted to the Ascend AI Processor, which can be used to verify the single-operator functionality.	Describes how to customize and modify models such as Caffe and TensorFlow models to make them compatible with ATC in scenarios where the ATC tool cannot be directly used.