Documentation

TensorFlow Model Development

Document

Description

Key Resource Acquisition

Environment Setup

Describes how to deploy the development and operating environment of the TensorFlow model on the Ascend platform.

Model Porting

Describes how to migrate Python training scripts based on TensorFlow 1.15 or TensorFlow 2.6.5 to the Ascend platform for execution to achieve optimal precision and performance. The key content is as follows:

  • Automatic porting: The porting tool can automatically port native TensorFlow training scripts to those supported by the Ascend platform. For APIs unportable by the tool, modify your training scripts according to the tool report.
  • Manual porting: Algorithm engineers need to manually analyze the API support of TensorFlow training scripts by referring to the document and modify the APIs so that the scripts can be executed on the Ascend platform. This method is complex..
  • Model training: Perform single-device or distributed training on the Ascend platform for the script that is successfully ported.
  • Accuracy tuning: If the ported model can be commissioned on the Ascend platform but the accuracy does not meet the requirements or the convergence is poor, you can perform accuracy tuning by referring to this section.
  • Performance tuning: After the model is migrated, the functions of the model can be commissioned on the Ascend platform. If the performance does not meet the requirements, you can tune the performance by referring to this section.

TensorFlow Parser Scope Fusion Pattern Developer Guide

Provides guidance for developers to develop Scope fusion patterns of the TensorFlow framework. Multiple small operators in a scope can be replaced with a large operator or a combination of multiple operators to improve the execution efficiency of the TensorFlow computational graph on the Ascend AI Processors.

This document is intended for model developers. After reading it, you will be able to:

  • Understand the implementation scheme and method of scope fusion.
  • Develop additional scope fusion patterns based on the given samples.

TensorFlow Parser Scope Fusion Pattern Developer Guide

TensorFlow Parser Scope Fusion Patterns

Scope fusion is a scope-based fusion capability. Small operators in the scope are replaced with one larger operator or a composite of operators to improve efficiency.

Describes the built-in scope fusion patterns.

TensorFlow Parser Scope Fusion Patterns

API Reference

Describes the APIs provided by TF Adapter and TF Parser Scope fusion patterns.