Towhee：走进新一代机器学习平台

1 .2021-12-09 Leetcode Weekly Contest Towhee: State-of-the-art Computer Vision in 5 Lines of Code 

2 . 01 Challenges in Modern ML CONTENTS 02 Solving ML Gaps 03 Say Hello to Towhee 04 Call for Contributors 

3 .01 Challenges in Modern ML 

4 .Constantly reinventing the wheel • Dozens of BERT implementations on Github • Which implementation should I choose? 

5 .Difficult for beginners to get started • One architecture can result in many models • What model works best? 

6 .Running models at scale • DevOps teams have dedicated MLOps subteams • I have a model, now what? 

7 .Before Towhee After Towhee 

8 .02 Solving Gaps in Machine Learning 

9 . Application-oriented, not model oriented Make ML accessible in a couple lines of code Re-introduce the concept of pipelines and operators 

10 .03 Say Hello to Towhee Zilliz’s Application-Oriented Machine Learning Platform 

11 . An application-oriented machine learning platform, built on top of Pytorch and Tensorflow 

12 . Towhee hub Pipeline inputs Pipeline outputs ML pipeline 

13 .Pipeline 

14 .Graph 

15 .04 Call for Contributors 

16 . Towhee 1.0: 2021-12-31 • First stable general release • Image and music embedding pipelines Towhee 1.x • Optimized for single instance Future work • High-level pipeline creation • GPU support Towhee 2.x • Cloud native • Inter-machine optimization 

17 .Thanks for listening! towhee.io towhee-io/towhee