Open Source Collaboration (TensorFlow Dev Summit 2018)

Key Takeaways

[Music] hi everyone now I was gonna talk to you about my plans to reanimate dinosaurs with tensorflow but I don't want to steal those guys thunder so actually I'm here to talk about open source collaboration in the tensorflow project and that's my job at Google to work on growing the participation and the collaboration in the project in the whole community so you guys all here and everybody watching on the livestream are a huge part of this already if you saw a slide like this at the beginning of the day in Rogers keynote you'll have seen the numbers have already ticked up in fact in the five days I've been monitoring this slide I've kept increasing the numbers because the amount of participation on tensorflow is kind of staggering as an open-source project it blows my mind when I came to to work on it and so much of that is due to the participation of everybody here in the community there are parts of tensorflow that wouldn't exist and many of them without that collaboration for instance whether it's spark connectors whether its support for particular architectures and accelerators or maybe certain language bindings we don't only benefit from a huge amount of adoption by being open source but as the adoption grows this is the way we sustain and grow the project you saw this map earlier as well this is just some of the github stars that gave it their locations that we could map and respond authors Norway in far south as the Antarctic and what is really obvious right here now is although there's a large team at Google developing tensorflow there are far many more people outside and far many more places who using it and the truth of a popular open source project is the more adoption is more demand there's so much resort to do together to grow tensorflow now yeah that thing where you turn up to a party and like everyone's having a good time and they all seem to know what they're doing and why it's such fun but who am I gonna talk to and what are they all looking at and sometimes a large open-source project can be a bit like that you want to get involved in contributing to tensorflow but where do you start you think this module this feature over here is something you want to work on but who do you find where do you find the right person to talk to how do you learn you know what what we're thinking about the direction for this so we've heard some of those things from you and we recognize that we want to improve our openness a transparency and our participation we're trying to work on making easier to get involved in tensorflow we've already for instance refreshed our roadmap which you can find on the tensorflow website about the general direction of a lot of the code and we'll do that a lot more regularly but I want to talk about four initiatives that we have going that like labelers all to work together more effectively and faster the first of these is simple it's a central community for everybody who's working and contributing to tensorflow github has so much energy going on there's so much great debate in all the issues if you look in there and the pull requests really thoughtful conversations and contributions thank you for your important that and thank you for being part of that but what we don't have is a central place where we can collaborate so what were we bringing online in fact you can go and subscribe now is a mailing list developers tents Florida log and that's a place for contributors to tensorflow to work together so if you're contributing tensorflow or want to contribute towards tensorflow go and join and we can work together as a community to discuss design ideas to get feedback to coordinate together and many of the projects that you've heard about today have mailing lists that you can also find it tents Florida org slash community whether it's TF light or the tents of load-up j/s that's collaboration nobody talked about the fact that there are many use cases outside of Google that the core team just don't see in fact many more much more happens outside than inside the core team so we want to make it possible for people with shared interests in a project to work together this is where the beauty of open source comes in how do we do that so we're setting up a structure for groups to work on areas of tense flow together special interest groups or sakes we've been piloting the first of these for a few months now this is called cig build and it's all about building packaging and distributing tensile foam familiar with tensorflow you know we build it in a certain way well guess what not every architecture not every application finds that the best way for them Phenix for instance linux distributions want to build against the shared libraries in the distribution that starts something we do so it brought together a bunch of stakeholders over Linux distributions companies like Suzhou canonical' Red Hat companies like IBM Nvidia Intel to collaborate in a group that will look the build of tens flow and make that work more effectively for more people in future that's just one group the pilot we want to pave the cow paths so where there is energy and where there are people collaborating on a particular thing that's a great candidate for bringing a special interest group together this week we're also bringing online the group for tents aboard where key stake holders of the tents aboard ecosystem can focus on design and collaboration and also for the rust language bindings which is completely community driven part of tensorflow each sig is going to have a different personality different way of working a different community but the common thing is we're going to provide forums where if you have a shared interest intense flow in a particular area we can focus on it now I'd like to talk about the design of tensorflow you know one of the most amazing things and the benefits of tensorflow is that the code that we release is the code that Google uses on a daily basis it's kind of remarkable for an open-source project and so we're really careful about changes we're really careful about design we have a commitment obviously to the API through the 1-point x-series release and we have design reviews internally so when things change we have proposals we get feedback but by now you're thinking well you just said that so many use cases assembly users are outside of Google yet you're having these design reviews inside so what we're going to do is open up a public feedback phase to our design process so we engage much more broadly with every contributor every user about how that might affect their needs and what their opinions are keep an eye on the developers at tensorflow dog mailing lists for this because that's where I'll notify it coming online in the next couple months but my hope that this is that this process will also be a way that everybody can communicate and discuss in a known process about the future direction of tensorflow whether you're in the core team here at Google or in the broader community so contributing to tense flow isn't just about issues or PR requests in fact I would say I reckon that there's so much more probably more energy going into people who are blogging running meetups doing presentations teaching doing courses so many universities around the world and we want to amp up and support the amount of content that educates and highlights tensorflow and so we're really excited already that so many of you do such amazing jobs and we'd like to be able to point everybody in the tensorflow community to the work that you're doing so there's a couple things that were launched to support this the first you've probably already heard we now have a blog for tensorflow a blog but Center Florida org one of the things that I'm most excited about with this blog is it as well as important announcements and education that we can put out we're setting it up from the beginning to involve content from around the web and in the community that's one reason we're using the medium platform for this to make it easy for us to integrate content from around the web and still give you the credit for the work you've done so we'd really like to hear from you if you have a blog post an article you want to get into the tense flow publication get in touch secondly and if you're on the live stream watching us you've kind of found out about this we have a YouTube channel that's launched today now one of the things I'm most excited about and this is a show called tensorflow meets this is where we're able to get out into the world of contributors and users and really highlight some of the use cases highlight the work of everybody again this is a chance for you to we'd love to meet you we'd love to chat with you about what you're up to and have you featured on the YouTube channel again reach out to us we'd love you to be a part of it there is one URL to get involved in all these things that I mentioned tensa Florida org slash community so if anyone's mentioned the mailing list or a group are something to you about that today please go to that URL and you find the resources there if I hope that tensorflow is going to continue to be a party there may be one you can find yourself part of a lot sooner and have more fun please feel free to reach out to me there's my email address aw Jake I'm and talk to you about anything to do with your experiences and collaborating and open source and tensorflow I'd love to hear about it thank you so much [Applause] [Music] [Music]