Dr. Ibrahim Haddad, The Linux Foundation | Open Source Summit Europe 2022

This is texturing TV. Hey guys, we're back at the open source. Software's Summit in lovely Dublin Ireland and we're talking with Dr.

Ibrahim Haddad who's the new head of the pie torch Foundation, which is an arm of the Linux foundation and we'll explain what all that means as we go along here, but pytorch is a machine learning framework that's people are getting very excited about originally developed at our friends now called meta Welcome to the show. Thank you very much and welcome to the open source Summit awesome. So why is there such excitement about Pine torch specifically because there's a lot of different machine learning Frameworks that are out there.

But what is it about this one? That seems to be garnering more attention? So a pie torch as a project is really the largest projects in its category.

So when we started tracking the open source ecosystem about, you know, five years ago or so and even today there are about you know, 12 to 15 different Frameworks that do more or less kind of similar functionalities within that domain. And over time for the past few years meta has done really an amazing job in growing the community of the pythosh project. So when you look at the project, there are over 2,400.

Code contributors to it coming from thousands of organization and there are other 18,000 projects that use phytorch as a dependency for them. So that has been really an incredible increase in adoption and contribution to the project making it really kind of the star project within its category and a lot of people are excited to see the project move to the foundation transition to an open governance model. And then from this point on grow as a foundation project where meta and the other founding members are just one of several companies that are formally funding the efforts of the project and supporting it under the framework of the foundation.

Now, you would think that the folks who work on AI are some of the smartest guys in the world, but it seems like maybe we've been Reinventing the same wheel over and over again in these different Frameworks. So well this whole process ultimately get more efficient Yeah, so it's really an interesting question because you see the same situation in many other areas within the eye domain or even within the cloud domain or any other technology domain where you have a lot of effort that look like they're doing the same thing. There's a lot of fragmentation and over time efforts consolidate and then you have at the end of the day, you know a few years later one or two winners that grab most of the market share.

So when we look at the domain of AI You know five years ago, there were about you know, 70 key projects in the ecosystem that we identified as critical that we were following up on and featuring. And what's really interesting is even though you know that landscape captured about so many projects today. We are about 330.

There's a lot of fragmentation and a lot of the project trying to do more or less the same thing and that goes part to the fact that a lot of these efforts were internal efforts, you know, there were like a feature in a product or they were an internal effort to achieve acts why or Z and over time these organization that founded these efforts realized that the value is not in the library or creating or it's not in the platform or it's not in the framework and they decided to open source these and build the communities around it and use it as a foundation for what they're building the real value for this such organization is in the data and the models that need these Frameworks to operate on so that's how we ended up with, you know, 12 to 15 deep learning framework with you know, you know 15, you know machine learning framework with you know, 20 learning framework or training. Morgan so on and over time really the market in this case the market are the adopties of this project and contributors. They decide who's the winner based on you know, how friendly the community is how easy it is to use the software how useful it is feature set and many other factors and you know, we end up with, you know, couple two three major players.

so in the fighters situation This is really kind of a monumental. Move really first it is a major commitment from meta to open source, because if you have such an amazing project, it's been widely adopted and massive community and there's you know hundreds of you know tens of thousands of organizations adopting it, you know, it is really you're going out of your way to move it to a foundation. I mean, you can continue down that part but you know matter team and meta AI team decided hey, you know, we want this to be even more impactful are much Global level and we want to take that project move it and transition it into a foundation and we as Mada for instance become one of several players that are funding the effort and ushering it into it's kind of the next future Evolution.

So really I'm super excited about this and you know, if you ask me three years ago, I wouldn't have guessed it's gonna happen and in Ai and data which I run Alafaya and data. You know every month we would have our calls and we discuss you know, how are we gonna bring pipes to the foundation? And you know, it happened on its own, you know, there's a time where the organization realizes.

Hey, this is something good we need to do for our community and this shows kind of a certainly extreme High maturity and strategy with open source, and then you know that and here we are. Are we on the cusp of? Getting to a point where just about every application you're going to build going forward is going to have some kind of AI machine learning capability built into it seems like it's becoming core capability.

And as that progresses does the standardization of the Frameworks help accelerate that transition because people won't be sitting there trying to figure out which of these 13 things to use. Yeah. So I think there are two two parts to your question.

So on the first part, absolutely, I think as we kind of move forward Everything today we do is touch by AI, you know, whether we know it or not, you know, when you go to say Netflix all these different suggestions coming on the screen are based on previous things on different models that run and try to get your preferences or when you go to the Amazon and you want to buy products it it's just gives you suggestion and this is just kind of really kind of the tip of the iceberg and there are a lot of things happening today in the insurance sector in the banking sector where it's a lot of it is driven by Ai and machine learning. So I think going into the future. This is here to stay that's kind of the future is a lot of products services in anything we touch is Really Gonna Be by some part driven and enabled by AI now to your your kind of next question.

I think you know with respect to Pi torch. eventually All the adoption will have to filter down. So when we look at the Deep learning Frameworks, for example, I know for instance in China.

Angel is a deep learning framework that is extremely widely adopted in China and they have thousands of organizations using it and they have you know, a few dozen or large organizations contributing to it, but it's kind of very limited and it's adoption to that geography. And there are multiple such instances and I think when you are in organization and you want to adopt open source technology, you typically want to have a project that is awesome. Any foundation where there isn't a single control point on the project where the foundation manages the project it owns its it infrastructure.

It controls. It's assets. It owns its intellectual property.

So it's never in any way shape or form ever use against you as an adoptee for whatever reason so when you find a project that meets your needs It is hosted in the foundation it offers you a path to become a leader in the project. If you invest in it in terms of resources and contributions and that project has its assets sitting in a neutral home that is vendorfree from all the different pressures. Then this is additional motivations for you say, you know, I want to do that.

I want to put my money and my resources and rely on this specific open-source project in my product or service because I know it's gonna be there in five years. I know there's the foundation looking after this project nurturing its Community supporting is expenses and so on so it's certainly provides a lot of incentives for organizations. Will we see more of a democratization of these capabilities so that the average mere mortal developer or maybe it shows up in a low-code tool or a no-code tool somehow or other to invoke these capabilities or is it always going to sit up here in this kind of high priest data scientists kind of Realm or you know, or are we gonna see this technology kind of filter down through the ecosystem.

So it absolutely is gonna filter down and I think a great indication of this is the ecosystem around whether it's by torch or any kubernetes is a great example because you know, it's pretty Decor. I mean, it's they've been around for a few years. Now when you look at the ecosystem of kubernetes, there are hundreds of supporting Tools around kubernetes that helps people with their deployment and their monitoring and their security and their auditing and many many other aspects and similarly to that.

I think the future will be kind of Looks the same for pytorch or any of the other winners, you know within their own category where you have the major project which is pytors for example, and you have this growing ecosystem of tools and enabling Tools around it that will enable really anyway necessarily not an AI PhD expert or not a data scientist to be able to use these tools and rely on in order to generate insights and business intelligence for whatever purpose. They wanted you one of the challenges we hear a lot about is that the models today are built mainly by data scientists who are working on their timeline and they can be anywhere from three months to a year to build the thing and then there's a devops team that's building applications and they're doing updates. Yeah several times a week sometimes monthly and we're trying to like insert the models into these devops workflows and we don't really have the best practices to do that yet.

So is that something you guys are going to be thinking about is And I like make it easier to insert the models into the applications and kind of build all the inference engines and then update that because it's a whole cycle of things. Yeah, so good question. So on the short term, you know for the next few months as a foundation, you know Fighters Foundation.

We are putting a lot of effort and we're our kind of almost exclusive focus is going to be on the transition. So we we had really An amazing amount of work that's been done for the past five to six months by, you know meti IMD Nvidia and you know the other partners Microsoft and Google in pulling efforts, and we I think we were looking at or internal tool where we had over 2,000 tasks that needs to be accomplished to get us on the keynote stage on Wednesday to announce. So a lot of people were involved in a thousands of emails hundred of calls.

So the next few months are lots of lawyers and lots of efforts come in the coming weeks and months will be on the transitioning of the projects assets and transitioning the governance and you know, all of that and I think in parallel to that we are today going through the exercise of kind of pulling feedback bringing in feedback. You know, what do you think now that we have a foundation and we have an operating budget and we have staff and stuff that we're trying how we as a foundation should be supporting the work as a technical community that can be training and we've released one course and we have multiple on the roadmap that can be certification program better improved infrastructure and you know many other key areas and one of these key areas for instance is developer experience documentation and you know the point you brought in but I think we're a little bit not there yet in terms of roadmap as we kind of bring in and to kind of take a deep breath after the announcement and realize okay. Now we have like 5,000 more tasks to get us to the next phase.

But definitely all of the mentions of the items you mentioned are. Items, we are we kind of for our off and we realize that you know, we need to get there and the foundation just formalizes that effort as a community. But if I wanted to be a contributor or maintainer, there's already a large community there to participate in and I can join in there.

So just how big is that community and how does one become a member of it? So are you talking a community as a developer Community not kind of a pie torch membership. Well now yeah, not a vendor partner sponsor more just the average person.

Oh, so yesterday actually had a fire chat with so much intella Summit is a co-founder lead maintainer of the project was there from day one and he's an aid Searcher at matter and I actually ask him. So the same question you're asking me so I'm just gonna relay really the message and thank you to submit for you for that. So basically today there's actually a lot of documentation available and it's really I was really impressed by it as we started working with the partners that was meta.

I was really curious about, you know, learning more about the project and I started looking over a lot of the online documentation. So there's really very detailed documentation that whose goal is to support individuals. You know, I'm curious I'm a student.

I want to I'm studying AI or I'm a developer at whatever company and I want to explore if pytosh is suitable for my needs. So there's a lot of documentation available for you to go and start and you know be able to do the equivalent of hello world and see it work and then kind of get engaged more and more there's a contributor guidelines there's contributed documentation and we're actually working and have some ideas for additional things that we can do to enable developers and this is actually on our to do to-do list. So it may earnings just well, there may be a pie torch University someday, is that part of the equation so I don't know if this is gonna be a partridge University, but definitely you are keen on making available education resources.

We're starting making courses available via the next foundation training department and we have at least one more course. Being worked on today as we speak and it is something that we consider high priority. We're making education on the framework itself highly accessible as you know, as projects grow and grow it becomes really difficult for people to come in and engage because there's this intimidation Factor.

Wow, there's like hundreds of thousands. If not more lines of code a lot of modules a lot of different kind of supports where to get started. So we want to make the developer Journey extremely easy and and kind of flowing, you know, super easy for the developers.

We want to kind of remove any friction and the way to do that is providing training documentation, and we're even considering certification similar to the kubernet word where you have developers certifications and certification for organizations and so on so and this is really good not have been possible without met us commitment to open source and to moving Listen to foundation and enabling the contributions of others and you know having a foundation care for the project. In terms of people becoming members of the foundation itself, whether it's gold silver. Are you looking there bringing more and user organizations?

Because there's a lot of vendors that are carry in this space. But sometimes I feel like the end user organizations Trail a little bit when they wait to see how things are shaking out but I mean are there more end user organizations involved or do you want them involved? And is that something that's a goal?

Yes. So I think at this point we only announced the founding members you can imagine the amount of work that went into kind of getting us here. So we didn't want during that formation days or months to be stuck in a way in discussions about you know, how many lovers should we have?

What are the membership fees that will help us sustain the foundation what are the benefits associated with you know, these are discussions that can really drunk on and maybe take us off our goal to announce. So we decided that's just focus on the founding level and then as soon as we convene as a board after Announcements which will be in a couple weeks. We're gonna have a lot of items to discuss and one of them is membership.

You know, what are the levels we need to have and you know, I think the personally I believe the user group are extremely beneficial. If you look at the cncf they have that and it's provided, you know in immense amount of feedback back into the foundation into the project and it seem useful. So I think there are really a lot of opportunities, you know, whether it's the kind of the end user forums if you wish or cold or you know pie torch Champions, you know, if you think of cncf Champions or ambassadors, so there are really a lot of really positive and successful models looking at other umbrella Foundation that we can look at learn from them and kind of replicate within the restrictions or you know, the model of Fighters All right.

So let's get out our AI augmented crystal ball. Where are we going to be a year from now? We're gonna be in a much better place.

We're gonna be you know, more companies more developers and this is not kind of being overconfident. It's just like when we look back at the growth of pytors for the past, you know, five or six years. It's been really Monumental and year over year more organizations are helping the project more organizations are relying on it more projects using it as dependency and more developers contributing to it.

And we actually monitor a couple things we monitor the Contributions from the same developers. So same developers are contributing more and exists and new developers are engaging on a long-term basis. So we're actually in the process of onboarding pytours to the foundation and adding it to our LFX platform, which will give us much deeper insights into all developer related contributions to the project.

So I think in a couple months if we sit back and you ask me the same questions, I'm gonna be able to pull up a slide and say Hey, you know a month ago. We were at this in terms of contribution number of commits adopties Etc. And today we're at this and I'm really confident in how the project is moving that we're on a post of Trend in terms of all these different metrics.

All right, folks write this thing down because in a few years, you're gonna sit back and go. I remember when this thing just got started. Awesome.

Thanks for being here. Thank you guys will be back in a few minutes.