Introduction

TPU-MLIR is a TPU compiler dedicated to processors. This compiler project offers a complete toolchain that can convert various pre-trained neural network models from different deep learning frameworks (PyTorch, ONNX, TFLite, and Caffe) into efficient model files (bmodel/cvimodel) for operation on the SOPHON TPU. Through quantization into different precisions of bmodel/cvimodel, the models are optimized for acceleration and performance on the SOPHON computing TPU. This enables the deployment of various models related to object detection, semantic segmentation, and object tracking onto underlying hardware for acceleration.

This course is mainly divided into three parts:

  1. Building and configuring a local development environment, understanding related SOPHON SDK, TPU-MLIR compiler core theories, and relevant acceleration interfaces.
  2. Converting and quantizing example deep learning models from ONNX, TFLite, Caffe, and PyTorch, along with methods for converting other deep learning framework models into the intermediate ONNX format.
  3. Guiding participants through the practical porting of four instance algorithms (detection, recognition, and tracking) for compilation, conversion, quantization, and final deployment onto the SOPHON 1684x tensor processor's TPU for performance testing.

This course aims to comprehensively and visually demonstrate the usage of the TPU-MLIR compiler through practical demonstrations, enabling a quick understanding of converting and quantizing various deep learning model algorithms and their deployment testing on the SOPHGO computing processor TPU. Currently, TPU-MLIR usage has been applied to the latest generation deep learning processors BM168X and CV18XX developed by SOPHGO, complemented by the processor's high-performance ARM core and corresponding SDK for rapid deployment of deep learning algorithms.

Advantages of this course in model porting and deployment:

1. Supports multiple deep learning frameworks

Currently supported frameworks include PyTorch, ONNX, TFLite, and Caffe. Models from other frameworks need to be converted into ONNX models. For guidance on converting network models from other deep learning architectures into ONNX, please refer to the ONNX official website: https://github.com/onnx/tutorials.

2. User-friendly operation

Understanding the principles and operational steps of TPU-MLIR through the development manual and related deployment cases allows for model deployment from scratch. Familiarity with Linux commands and model compilation quantization commands is sufficient for hands-on practice.

3. Simplified quantization deployment steps

Model conversion needs to be executed within the docker provided by SOPHGO, primarily involving two steps: using model_transform.py to convert the original model into an MLIR file, and using model_deploy.py to convert the MLIR file into bmodel format. The bmodel is the model file format that can be accelerated on SOPHGO TPU hardware.

4. Adaptable to multiple architectures and modes of hardware

Quantized bmodel models can be run on TPU in PCIe and SOC modes for performance testing.

5. Comprehensive documentation

Rich instructional videos, including detailed theoretical explanations and practical operations, along with ample guidance and standardized code scripts, are open-sourced within the course for all users to learn.

SOPHON-SDK Development Guide https://doc.sophgo.com/sdk-docs/v23.05.01/docs_latest_release/docs/SOPHONSDK_doc/en/html/index.html
TPU-MLIR Quick Start Manual https://doc.sophgo.com/sdk-docs/v23.05.01/docs_latest_release/docs/tpu-mlir/quick_start/en/html/index.html
Example model repository https://github.com/sophon-ai-algo/examples
TPU-MLIR Official Repository https://github.com/sophgo/tpu-mlir
SOPHON-SDK Development Manual https://doc.sophgo.com/sdk-docs/v23.05.01/docs_latest_release/docs/sophon-sail/docs/en/html/

Chapters （ 22Lesson）

1_ introduction
Start Learning
1.1 Background and explanation of terms
To do
Start Learning
1.2 SOPHONSDK Package Profile Guide
To do
Start Learning
2_ Overview of the Compiler
Start Learning
2.1 Introduction to TPU-MLIR Compiler
To do
Start Learning
2.2 Overall Architecture Design of TPU-MLIR
To do
Start Learning
2.3 Overview of the theoretical core of TPU-MLIR
To do
Start Learning
2.4 Development Interface Guide
To do
Start Learning
2.5 Advantages of the Deep learning compiler TPU-MLIR
To do
Start Learning
3_ Development Environment
Start Learning
3.1 Hardware and software requirements
To do
Start Learning
3.2 TPU-MLIR development environment and commonly used dependency packages
To do
Start Learning
3.3 Build cross-compile environment and validation
To do
Start Learning
4_ TPU-MLIR Quickstart
Start Learning
4.1 TPU-MLIR-compiled ONNX models
To do
Start Learning
4.2 TPU-MLIR-compiled Torch models
To do
Start Learning
4.3 TPU-MLIR-compiled TFLite models
To do
Start Learning
4.4 TPU-MLIR-Compiling Caffe Models
To do
Start Learning
4.5 Other Deep Frame to Intermediate Format ONNX Guidelines
To do
Start Learning
4.6 Mixing Accuracy Usage
To do
Start Learning
4.7 Pre- and post-treatment with TPU (BM168X)
To do
Start Learning
4.8 Pre-treatment with TPU and CV18XX Processor Utilization Guide (CV18XX)
To do
Start Learning
5_ Algorithm deployment in practice
Start Learning
5.1 Porting and Testing of Text Recognition Algorithm Based on PP_OCRv2
To do
Start Learning
5.2 Porting and Testing of CenterNet-based Image Segmentation Algorithm
To do
Start Learning
5.3 Planting and Testing of YOLOv5-based Target Detection Algorithm
To do
Start Learning
5.4 Planting and Testing of YOLACT-based Target Tracking Algorithm
To do
Start Learning

Objective

After completing this course, students will be able to master the following capabilities:

  • Have a comprehensive understanding of the conversion and quantization process of the TPU-MLIR Deep learning compiler.
  • Master the compilation, conversion, and quantization of models from popular deep learning frameworks such as ONNX, Torch, TFlite, and Caffe.
  • Be familiar with converting models from other deep learning frameworks into the intermediate format ONNX and complete model compilation, conversion, and quantization.
  • Understand the interface invocation and compilation usage process of the SOPHON TPU inference library SDeep learningL.
  • Learn to use the TPU-MLIR Deep learning open-source compiler for model compilation, conversion, quantization, porting deployment, and optimization.

Course Participants

Deep learning enthusiasts or open-source community developers with a certain foundation in deep learning and Python development, familiarity with Docker, and Linux operations. Users can contribute to the improvement of this course while using it, collaborating with SOPHGO to build an advanced Deep learning compiler for the era.

Course Recommendation

course-cover

Compiler development

As a bridge between the framework and hardware, the Deep learning compiler can realize the goal of one-time code development and reuse of various computing power processors. Recently, Computational Energy has also opened source its self-developed TPU compiler tool - TPU-MLIR (Multi-Level Intermediate Representation). Tpu-mlir is an open source TPU compiler for Deep learning processors. The project provides a complete tool chain, which converts the pre-trained neural network under various frameworks into a binary file bmodel that can operate efficiently in TPU to achieve more efficient reasoning. This course is driven by actual practice, leading you to intuitively understand, practice, and master the TPU compiler framework of intelligent Deep learning processors.

At present, the TPU-MLIR project has been applied to the latest generation of deep learning processor BM1684X, which is developed by Computational Energy. Combined with the high-performance ARM core of the processor itself and the corresponding SDK, it can realize the rapid deployment of deep learning algorithms. The course will cover the basic syntax of MLIR and the implementation details of various optimization operations in the compiler, such as figure optimization, int8 quantization, operator segmentation, and address allocation.

TPU-MLIR has several advantages over other compilation tools

1. Simple and convenient

By reading the development manual and the samples included in the project, users can understand the model conversion process and principles, and quickly get started. Moreover, TPU-MLIR is designed based on the current mainstream compiler tool library MLIR, and users can also learn the application of MLIR through it. The project has provided a complete set of tool chain, users can directly through the existing interface to quickly complete the model transformation work, do not have to adapt to different networks

2. General

At present, TPU-MLIR already supports the TFLite and onnx formats, and the models of these two formats can be directly converted into the bmodel available for TPU. What if it's not either of these formats? In fact, onnx provides a set of conversion tools that can convert models written by major deep learning frameworks on the market today to onnx format, and then proceed to bmodel

3, precision and efficiency coexist

During the process of model conversion, accuracy is sometimes lost. TPU-MLIR supports INT8 symmetric and asymmetric quantization, which can greatly improve the performance and ensure the high accuracy of the model combined with Calibration and Tune technology of the original development company. In addition, TPU-MLIR also uses a lot of graph optimization and operator segmentation optimization techniques to ensure the efficient operation of the model.

4. Achieve the ultimate cost performance and build the next generation of Deep learning compiler

In order to support graphic computation, operators in neural network model need to develop a graphic version; To adapt the TPU, a version of the TPU should be developed for each operator. In addition, some scenarios need to be adapted to different models of the same computing power processor, which must be manually compiled each time, which will be very time-consuming. The Deep learning compiler is designed to solve these problems. Tpu-mlir's range of automatic optimization tools can save a lot of manual optimization time, so that models developed on RISC-V can be smoothly and freely ported to the TPU for the best performance and price ratio.

5. Complete information

Courses include Chinese and English video teaching, documentation guidance, code scripts, etc., detailed and rich video materials detailed application guidance clear code script TPU-MLIR standing on the shoulders of MLIR giants to build, now all the code of the entire project has been open source, open to all users free of charge.

Code Download Link: https://github.com/sophgo/tpu-mlir

TPU-MLIR Development Reference Manual: https://tpumlir.org/docs/developer_manual/01_introduction.html

The Overall Design Ideas Paper: https://arxiv.org/abs/2210.15016

Video Tutorials: https://space.bilibili.com/1829795304/channel/collectiondetail?sid=734875"

Course catalog

 

序号 课程名 课程分类 课程资料
      视频 文档 代码
1.1 Deep learning编译器基础 TPU_MLIR基础 √ √ √
1.2 MLIR基础 TPU_MLIR基础 √ √ √
1.3 MLIR基本结构 TPU_MLIR基础 √ √ √
1.4 MLIR之op定义 TPU_MLIR基础 √ √ √
1.5 TPU_MLIR介绍（一） TPU_MLIR基础 √ √ √
1.6 TPU_MLIR介绍（二） TPU_MLIR基础 √ √ √
1.7 TPU_MLIR介绍（三） TPU_MLIR基础 √ √ √
1.8 量化概述 TPU_MLIR基础 √ √ √
1.9 量化推导 TPU_MLIR基础 √ √ √
1.10  量化校准 TPU_MLIR基础 √ √ √
1.11 量化感知训练（一） TPU_MLIR基础 √ √ √
1.12  量化感知训练（二） TPU_MLIR基础 √ √ √
2.1 Pattern Rewriting TPU_MLIR实战 √ √ √
2.2 Dialect Conversion TPU_MLIR实战 √ √ √
2.3 前端转换 TPU_MLIR实战 √ √ √
2.4 Lowering in TPU_MLIR TPU_MLIR实战 √ √ √
2.5 添加新算子 TPU_MLIR实战 √ √ √
2.6 TPU_MLIR图优化 TPU_MLIR实战 √ √ √
2.7 TPU_MLIR常用操作 TPU_MLIR实战 √ √ √
2.8 TPU原理（一） TPU_MLIR实战 √ √ √
2.9 TPU原理（二） TPU_MLIR实战 √ √ √
2.10  后端算子实现 TPU_MLIR实战 √ √ √
2.11 TPU层优化 TPU_MLIR实战 √ √ √
2.12 bmodel生成 TPU_MLIR实战 √ √ √
2.13 To ONNX format TPU_MLIR实战 √ √ √
2.14 Add a New Operator TPU_MLIR实战 √ √ √
2.15 TPU_MLIR模型适配 TPU_MLIR实战 √ √ √
2.16 Fuse Preprocess TPU_MLIR实战 √ √ √
2.17 精度验证 TPU_MLIR实战 √ √ √
course-cover

Milk-V Duo Development Board Pratical Course

This course introduces the hardware circuit design and basic environment set up, as well as provides some simple development examples and some basic Deep learning examples.

Milk-V Duo is an ultra-compact embedded development platform based on CV1800B. It has small size and comprehensive functionality, it is equipped with dual cores and can run linux and rtos systems separately, and has various connectable peripherals.

  • Scalability: The Milk-V Duo core board has various interfaces such as GPIO, I2C, UART, SDIO1, SPI, ADC, PWM, etc.
  • Diverse connectable peripherals: The Milk-V Duo core board can be expanded with various devices such as LED, portable screens, cameras, WIFI and so on.

Course features:

  • The content materials are rich and complete, including development board hardware design, peripheral interface instructions, basic environment set up method, and sample code scripts.
  • The learning path is scientifically reasonable, starting from the introduction and basic usage of the development board, and then leading to pratical projects to fully utilize the development board and provide reference for users' own development.
  • The pratical projects are rich, and the course provides many examples of practical code usage and function demonstrations. Different functions can be implemented by simply modifying and combining the code.

course-cover

SE5 Development Series Course

The deep neural network model can be trained and tested quickly and then deployed by the industry to effectively perform tasks in the real world. Deploying such systems on small-sized, low-power Deep learning edge computing platforms is highly favored by the industry. This course takes a practice-driven approach to lead you to intuitively learn, practice, and master the knowledge and technology of deep neural networks.

The SOPHON Deep learning microserver SE5 is a high-performance, low-power edge computing product equipped with the third-generation TPU processor BM1684 developed independently by SOPHGO. With an INT8 computing power of up to 17.6 TOPS, it supports 32 channels of Full HD video hardware decoding and 2 channels of encoding. This course will quickly guide you through the powerful features of the SE5 server.  Through this course, you can understand the basics of Deep learning and master its basic applications.

Course Features

1. One-stop service 

All common problems encountered in SE5 applications can be found here.

 • Provide a full-stack solution for Deep learning micro servers

 • Break down the development process step by step, in detail and clearly

 • Support all mainstream frameworks, easy to use products

2. Systematic teaching 

It includes everything from setting up the environment, developing applications, converting models, and deploying products, as well as having a mirrored practical environment.

• How is the environment built? 

• How is the model compiled? 

• How is the application developed? 

• How are scenarios deployed?

3. Complete materials

The course includes video tutorials, document guides, code scripts, and other comprehensive materials. 

• Rich video materials 

• Detailed application guidance 

• Clear code scripts 

Code download link: https://github.com/sophon-ai-algo/examples

4. Free cloud development resources 

Online free application for using SE5-16 microserver cloud testing space 

• SE5-16 microserver cloud testing space can be used for online development and testing, supporting user data retention and export 

• SE5-16 microserver cloud testing space has the same resource performance as the physical machine environment 

Cloud platform application link: https://account.sophgo.com/sign_in?service=https://cloud.sophgo.com&locale=zh-CN

Cloud platform usage instructions: https://cloud.sophgo.com/tpu.pdf