<aside> 💡 PyTorch NPU code conversion and training - the information on this page are condensed from the list of resources you can find below.

</aside>

Ascend PyTorch Resources

Table of Content

Background

PyTorch remains one of the most-used framework for developing ML&AI models

• PyTorch Packages layout

• PyTorch vs TensorFlow (Dynamic vs Static graph) - PyTorch is more Pythonic than TensorFlow and has an object-oriented approach while TensorFlow has several options from which you may choose

• PyTorch on Ascend Processor: PyTorch Adaptor - a patch to original PyTorch and use online solution to adapt PyTorch to run on Ascend AI Processors.

• 🛑PyTorch Adaptor Limitations🛑

• 🔍 Monitor NPU Resources

• [Situational] PyTorch Installation

Model Porting & Code Conversion

• Model Porting Pre-Check; (Adapted Operator List + How to Develop Operators)
• Model Porting Details

Special Topic: Distributed Training

• API Explanation