"Coach, I want to play basketball" -- AI Learning Series booklet for students who are making systems

 

Recently read Jeff Dean A new work of Pathways , Once again, such confusion appeared in my heart ： From the perspective of frontier work , System work and algorithm model work have long been combined with each other 、 The state of being indispensable . But in recent years, in the process of open source projects , And developers 、 Communicate with users , The biggest experience is that system engineers and Algorithm Engineers are almost always far away . The students look at each other , Probably “ It is used to wash herbs and refine pills ” vs “ Who makes great efforts to perform miracles and swing a sledgehammer ”……

The students who do the system want to follow up AI Technology trends , Swing the sledgehammer into the heart of the arithmetic students , Not easy . Almost all learning materials have a strong smell of scientific research , Too many formulas and theories , Too few projects and systems .

Born as a system AI lovers , We're doing it  [Towhee] (https://github.com/towhee-io/towhee) And  [Milvus] (https://github.com/milvus-io/milvus) In the process of the two open source projects, I have been waiting for a system tailored for people AI course , But there has been no such work .

In the near future , We plan to make a try by ourselves , Make our own system AI Study the tutorial . The project orientation is open source , Pure system angle , to ground , interesting . There are three series in the current plan , Focus on application and actual combat respectively 、 Inference system optimization 、 Cutting edge system technology ：

1、 Build from scratch AI application

• application CLIP Build text - Image cross modal search service

• The romance of technology house ： Picture and video quadratic ！

• Large scale map search system

• Cross modal video semantic retrieval

• Music copyright detection practice （ Planning ）

• Question answering system （ Planning ）

• No one knows how to browse efficiently better than I do Github ： by Github Build an open source project retrieval and recommendation system （ Planning ）

• The romance of technology house ： Your goddess always lacks a red trumpet ！（ Planning ）

2、 Build your own high-performance reasoning service

• First time to know TorchScript ,ONNX ,TensorRT

• be based on Nvidia Toolkit to speed up your reasoning service

• Towhee Reasoning service construction practice （ Planning ）

3、AI Intensive reading of system papers

• Pathways: Asynchronous Distributed Dataflow for ML

The next series of articles we will meet with you is  《 Build from scratch AI application ： application CLIP Build text - Image cross modal search service 》, Expected to be released this week . In this series , Will learn ：

• CLIP The basic working principle of cross modal search

• Text - Prototype construction of image cross modal search

• Advanced 1： Deploy vector database , Large scale vector recall

• Advanced 2： Deploy inference Services , Greatly improve reasoning performance

• Advanced 3： Vector data compression

To maximize the flow of the tutorial , We are already  [Towhee] (https://github.com/towhee-io/towhee) A series of preparatory work has been completed in the project , Include CLIP Integrated packaging of ,python On the side pipeline Programming interface (API) , Close to production practice docker Mirror image , Dry cargo is full. , Coming soon ！

For more project updates and details, please pay attention to our project  https://github.com/towhee-io/towhee , Your attention is a powerful driving force for us to generate electricity with love :)

---

Zilliz  With the vision of redefining Data Science , Committed to building a global leading open source technology innovation company , And unlock the hidden value of unstructured data for enterprises through open source and cloud native solutions .

Zilliz  To build the  Milvus  Vector database , To speed up the development of the next generation of data platform .Milvus  The database is  LF AI & Data  The foundation's Graduation Program , Able to manage a large number of unstructured data sets , In new drug discovery 、 Recommendation system 、 Chat robot has a wide range of applications .