当前位置：网站首页>Zhongshanshan: engineers after being blasted will take off ｜ ONEFLOW u

Zhongshanshan: engineers after being blasted will take off ｜ ONEFLOW u

2022-06-26 04:49:00 【ONEFLOW deep learning framework】

Zhongshanshan , First class technical engineer （ Internship ）, Graduated from Sun Yat sen University, majoring in information management and information system , He is now a quasi graduate student of the school of computer science, Sun Yat sen University .

During the period of university , She is the principal of several projects , And won many awards in University competitions such as data-driven innovation research , Accumulated technical practical experience . After graduation, , On the recommendation of senior brothers and sisters , She came OneFlow Internship .

She usually likes to play the guitar , He said he didn't play very well , But the noise and confusion made her feel very free 、 happy 、 And relax . She seems a little casual , In my senior year , She put GPA The ranking has reached No. 1 in major , But at this time, the scholarship was no longer evaluated ; She chose the school of computer science, Sun Yat sen University , It's because it won't be too tired to move the dormitory ……

The following is Zhong Shanshan's autobiography .

The Department setup of Sun Yat sen University is quite interesting , school of computing 、 School of software 、 School of electronics and information 、 School of Intelligent Engineering 、 Artificial intelligence colleges and so on AI The specialty of the direction . among , AI college is located in Zhuhai Campus , Therefore, I prefer to study in the computer college in Guangzhou , In this case , Don't be too tired to move the dormitory .

In my senior year , I put GPA Ranked first , But at this time we have not selected the scholarship …… In learning , I think it is very important to take more notes . Of course , Not everything in the textbook should be written down , But to dig deeply into the points of interest in the course , Share more in class , Find more teachers to discuss .

For me, , The quickest way to learn something is to use it to learn it , Go straight ahead . Many of my programming skills were also learned during my undergraduate years , Such as JAVA、Matlab etc. , I did R Language , Anyway, I learned a lot of things , But the real acquisition of programming skills depends on practice and interest .

For example, participate in “ Discovery cup ” In the software design contest , Quick back-end knowledge to complete the project . The most interesting ones are some self - written gadgets , Take reptiles for example , When I was just getting started, I found a website that looked simple , In the process of climbing, if you encounter problems, you can solve them , I learned a lot . If you want to advance , Find a website that doesn't look easy to climb , etc. IP When it was sealed , You have to find another way , Try to fill the hole you dug ：）

I also developed an autism warning model , Semantic analysis of autism behavior data of Baidu Post Bar autism bar , Extract typical features of autism , Achieve autism early warning , I am mainly responsible for data crawling 、 Feature Engineering and wechat applet construction . The feature engineering part is based on Python Realization RNN、TextRank、Word2vec, Carry out multi-layer filtering and semantic segmentation on Post Bar data .

I started with the recommended algorithm AI Of , The pure algorithm made it difficult for me to understand how this thing would land at the beginning , So I went to work on a project for a period of time . After understanding the basic structure and process of the project , I put my focus back on the algorithm project , My graduation thesis is to do NLP Research in . And in the OneFlow During the internship , I came into contact with something new —— The realization of operators , This gives me a deeper understanding of how operators work , When testing the model performance, you won't have no way to start .

If students want to improve their abilities , I think we need to start and finish , There will always be gains if you stick to it , I just grew up in the process .

And OneFlow From word of mouth, I got to know elder martial brothers and sisters , It is said to be the light of China , But compared with other domestic open source frameworks ,OneFlow It seems that the propaganda strength of "making a fortune through silence" ？

OneFlow Can practice remotely , It brings great convenience to interns in different places . actually , Due to the epidemic situation in recent years , Many Internet companies have successfully practiced telecommuting , No more commuting 、 Food and other problems , And as long as the work is properly arranged , You can balance your life and work .

In the early stage, I worked in the cloud platform group , stay Brother Chi Under the leadership of AI-Writer project , Then came the framework group , At present, under the guidance of Xiaoyu, the operator is developed and maintained .

Just arrived OneFlow when , I couldn't understand 《 How to be in OneFlow Newly added User Op》 This post , But when I imitate Xiaoyu's PR After implementing an operator , Look back at this article , I thought it was very well written . I also learned to use MkDocs and GitHub Build a website （https://zhongshsh.github.io/BlogInOF）, Often will be important PR The completion process of 、 The experience of stepping on the pit is sorted into the content of the web page , So that it can be retrieved and used when necessary .

It is a very happy thing to practice here , Telecommuting makes me less worried about commuting , The idea of open source makes me feel the power of everyone working together . meanwhile ,OneFlow There is a strong learning atmosphere , The success of brother Chi's night school made me crash C++, Send a meeting to Xiaoyu , He can explain patiently for more than half an hour , After all , This is a shake 6 I will never forget the M - magnitude earthquake debug The ox man of .

For work reasons , I have used OneFlow, Also involved in OneFlow Development and maintenance of . I wondered ,OneFlow Why and PyTroch As like as two peas （OneFlow Programming API Fully compatible with PyTorch, Common deep learning models only need to modify one line import oneflow as torch You can put it in PyTorch The model written in OneFlow Run up ）, But when I started to understand its underlying implementation , I was amazed again API The magic speed behind the same .

at present , Most people in academic circles still use PyTorch, It's really easy to use , And because the community is big , It is easy to find solutions to some problems . And the former industry giant TensorFlow The share of has been nibbled away step by step .OneFlow If you want to achieve PyTorch Height , The training of community users is essential .

stay OneFlow Internship , Many things are learned from scratch , It's very challenging for me , Especially when you finish something PR The process is very difficult , I knew I might get hammered , But it also means , My personal ability will take off by rocket .

Everyone else is watching

Welcome to experience OneFlow v0.7.0：GitHub - Oneflow-Inc/oneflow: OneFlow is a performance-centered and open-source deep learning framework.https://github.com/Oneflow-Inc/oneflow/