 Hello everyone, I'm Chupung Hong from Huawei and I'm super excited to have the opportunity to give this talk at OSS Summit this year for OSPOCon. Hopefully our topic of sharing our experience of running Mindsport open source community can help you. Now first of all, a bit self introduction. I'm currently the committee manager for Mindsport. I'm also serving as the tech member for LFA Foundation and Confidential Computing Consortium. So let me start with the general philosophy or the metaphysics that we adopt for our line of work regarding open source community operation. So in essence we borrowed a lot from RoboPresics theory of metaphysics of quality. So for our actually day to day work, we deploy something very similar to ML Ops to what data scientists use in AI. So for a typical machine learning operation pipeline, you will have the gather of the requirements and then you do the preparation for data. And then you perform training and do evaluations, see the result and then do a couple rounds of fine tuning until the result is very satisfiable and then you can deploy your model. So this is a typical pipeline for ML Ops. The community operation or we call the COM Ops we use has a very similar pipeline to ML Ops. We also have start with the requirements gallery. And then we have to define the KPIs or the key goals for our work to meet those requirements. After that, we identify the minimum viable product that we can deliver and we will try to deliver as fast as possible. So in the first round of reiteration, we do evaluation, see if the result is meeting our expectation, and we can do some fine tuning, and then we do what we call the actually a reproductive session. This is the OSPOR structure in action for our daily work. Okay, let me go into detail of how we do the community operation for the Minesport open source community. So first of all, what is Minesport? So Minesport is a very new cutting edge, awesome, all scenario AI framework. We open source Minesport just almost a year ago at the end of March last year. So similar to TensorFlow, PyTorch, MXNet, Minesport is also a deep learning inference and training framework. So this is a bird's-eye view of the Minesport OSPOR work just in one year. I think we have delivered an amazing result of building the most active AI open source community in China in just one year's time. So let me dive deep into the four aspects of our commops work in detail. First of all, communication through code. TinyMS is like an entry-level project that we developed for Minesport beginners. And we also created a crash course using TinyMS to help beginners and entry-level developers that want to learn deep learning but without the AI background to quickly get a hands-on. Another major aspect of our work is to do communication via content. Our team has created a ton of content, especially in the era of COVID. We are using a lot of the online streaming and also video conferencing to reach the developers. And also communication via outreach. We have been using a lot of social media for reaching out to developers. For example, we are using Slack for our global developer ecosystem. We created bootcamps. For example, last year we have the 21-day deep learning bootcamp. And we also organized hackathons, what we call the MINECOM Big Week. It's a bi-week event that goes across 12 or 15 cities in China for a hackathon. So the second aspect of the commops work is governance. Minesport, we believe, is among the first open source deep learning project that has an open governance structure. So we have a chartered governance model. We have the technical steering committee consists of 40 members that are across the globe. We have the special interest group and the working group in the community responsible for day-to-day discussion and the development. All of the SIG meetings and working group meetings and TS meetings are recorded and uploaded to various video sites to make sure that all of the discussion and the process are public and transparent. We also have put a lot of efforts on compliance for governance. And also diversity is on our mind constantly. Similar to the user group that many open source community have, we also have this structure called Minesport Study Group. So Minesport Study Group is designed for diverse, like local or topical or institutional social gatherings. So it has three types. First of all, we have the regional MSGs. As you can see, we have like 15 major cities in China have established MSGs and also across seven countries. For example, as you can see, this is the poster for the Russian MSG. And the second type is topical. We have a female developer president running the Women in Tech MSGs very successfully for about three gatherings. We also have other technical topics like Rusty AI. The third genre is institutional. So we have MSGs that target for, for example, for enterprise users to get involved in the community and how to best use Minesport. Or we have my MSGs that actually doing seminars in universities to help students and teachers getting familiar with Minesport technology. We also designed a lot of bot to help automate our infrastructure. For example, we have the CI bot running on both the GT code base and the GitHub code base. We have a sync bot built upon the GitHub action. The sync bot help mirror code in real time from GT to GitHub so that we have a two sync code base. We also have a DX bot that is the output from the developer experience. The DX bot help provide like suggestion of reviewers, suggestion of labels for your issues. And it actually improves a lot of things when developer interact with our CI bot. And at least we also use data-driven canvans heavily and, okay, open source collaboration. Minesport community has built a large collaboration among the major open source communities. We have a close collaboration with LFI Foundation. As you can see, this is one of the pictures I took for LFI Day in China. We also are working closely with this AIC Open Lab working group in Eclipse Foundation. We have a lot of technical collaboration with the SINSEF Foundation, with the Kubernetes project, and also with the Apache Foundation project like TBM. We also have a lot of the technical development collaborations. Okay. Another thing that we start doing this year, also we feel really excited is to open source the OSPO itself. As you know, although we are doing the operations for open source community, actually many of the operation, the side of operation is not open source. So this is why at the end of March this year, in the Open Atom Foundation, we created a new project called the Xerox Commerce Project. So what we believe is that maybe a mode of big tent could actually be a middle way where we have really great trained professionals doing the institutional work. And we're still working in a really freedom way like we have in the Bazaar. So this is what the Xerox Commerce Project also wants to achieve. Okay. This is the end of my talk. Thank you very much for listening in. Here are all the links of the MindSupport project. All of the code are open source and you are more than welcome to check it out and participate in the community and feel free to ask anything to submit an issue or submit a call request for our community. Thank you very much.