X3 Algorithm Toolchain

Sunrise3 (X3) processors are power-efficient and high-performance processors rolled out by D-Robotics in 2020. Wherein, X3 processors can be divided into X3M and X3E versions according to different application scenarios. X3 algorithm toolchain contains: Model Algorithm Processing and Embedded Model Prediction Library two important components. Model algorithm processing provides PTQ floating-point and fixed-point model conversion scheme, supports direct conversion of floating-point models obtained based on public frameworks such as TensorFlow, PyTorch or Caffe to fixed-point models. The resulting fixed-point model can be executed on the D-Robotics computing platform after being processed by the model compiler tool. The embedded model prediction library provides a series of support interfaces for completing inference using fixed-point models. For more details about the algorithm toolchain, please reading the subsequent contents.

To allow developers to quickly get started with the key steps of model training/conversion, deployment, validation, and inference, the components of the algorithm toolchain provides Sample packages as well as Docker images and model distributions to be used with the sample packages. These sample packages make the core business logic and key configuration parameters of each tool into scripts. The model release object is built-in with a large number of algorithmic models.The Docker image has a built-in development environment required to use the sample packages.

This document aims to provide you with a detailed introduction to the X3 algorithm toolchain related content. D-Robotics provides you with the following sections for your reading.

Catalog

• 1. Product Introduction
  • 1.1. OE Document Introduction
  • 1.2. OpenExplorer Introduction
  • 1.3. Toolchain Overview
• 2. Environmental Installation
  • 2.1. Environment Deployment
  • 2.2. Version Management of OE Package
• 3. Quick Start
  • 3.1. QAT and PTQ Introduction
  • 3.2. Algorithm Model PTQ + On-board Deployment Quick Start
• 4. Floating-point Model Convert to Fixed-point Model Guide
  • 4.1. Post-training Quantization (PTQ)
  • 4.2. Quantized Awareness Training (QAT)
• 5. Embedded Application Development (Runtime) Manual
  • 5.1. Embedded Application Development
  • 5.2. BPU SDK API DOC
  • 5.3. Basic Sample User Guide
  • 5.4. AI Benchmark User Guide
  • 5.5. Tool Introduction
• 6. Advanced Contents
  • 6.1. Efficient Model Design Guide
• 7. Appendix
  • 7.1. Model Performance Benchmark
  • 7.2. Model Conversion Toolchain Operator Support Constraint List
  • 7.3. Common Abbreviations
  • 7.4. Overview of X3
  • 7.5. Development Board Kit and System Image Description