Length : 1/2 day
This course introduces the Xtensa Neural Network Compiler (XNNC) v2.
The purpose of the Xtensa Neural Network Compiler (XNNC) is to convert a floating-point Convolutional Neural Network (CNN) into an optimized, fixed-point solution for Cadence Tensilica processors. The course aims to guide the user to the effective use of configuring and running supported neural network models through the different stages and modes of the XNNC workflow. This includes how to configure inputs and outputs for a network and the usage of the Custom Operator Application Interface (API).
The course describes the added support for Graph Lowering (Glow) Compiler techniques, Tensorflow convertors, Open Neural Network eXchange (ONNX), and Caffe2. The course also explains the use of new Quantization schemes and Xtensa Exchange Formats for generated DSP codes.
This course consists of four modules and labs:
- Introduction to XNNC v2
- Custom Operators
After completing this course, you will be able to:
- Configure and run the supported neural network models through different stages and modes of the XNNC work flow.
- Understand the use of new Quantization schemes and Xtensa Exchange Formats for generated DSP codes.
Software Used in This Course
Xtensa Neural Network Compiler (XNNC) v2.0 forWindows or Linux
Xtensa Software Tools release RI-2020.4, Xtensa Xplorer 8.0.13
Modules in this Course
- Module 01: About This Course
- Module 02: XNNC Inputs
- Module 03: XNNC Outputs
- Module 04: XNNC Customer Operations
- Software Engineers and Architects
- Neural Network Architects
- Machine Learning Researchers
You must have completed the following courses: