WEBVTT 00:00.000 --> 00:07.000 And in fact, there is much more than that. 00:07.000 --> 00:13.000 There is lack of transparency, but what you get when you get a model out of hugging face or something like this. 00:13.000 --> 00:22.000 And there is a lot of, you know, people don't really understand the terms that are with the licenses associated with those models. 00:22.000 --> 00:27.000 And so there is general, you know, there's people make mistakes or fun, but so. 00:27.000 --> 00:42.000 And so the model openness framework briefly, basically introduced ranking system to give transparency as to what's in model distribution. 00:42.000 --> 00:50.000 And we are looking at two different dimensions, it's very important to understand this because people often focus on the openness, but there is also completeness. 00:50.000 --> 00:53.000 So this is all based on the concept of open science. 00:53.000 --> 00:56.000 So let me get a little bit more into the details. 00:56.000 --> 01:04.000 When it comes to the completeness aspect, we are talking about what you actually get from this distribution. 01:04.000 --> 01:07.000 We have identified 17 different components. 01:07.000 --> 01:14.000 I won't go through them now, but there are three main types of components in a model distribution. 01:14.000 --> 01:19.000 There is code, but there is also data and documentation. 01:19.000 --> 01:25.000 So all of these participants in what's going to be in there and the completeness. 01:25.000 --> 01:35.000 On the openness side, it has to do with what under what terms those components, those components are made available to you. 01:35.000 --> 01:39.000 And so the model openness framework is very clear. 01:39.000 --> 01:45.000 You must have permissive licenses, no restriction whatsoever, right? 01:45.000 --> 01:49.000 And in fact, there is also another aspect which is often missed. 01:49.000 --> 01:54.000 We have, as I showed earlier, three different types of components. 01:54.000 --> 02:03.000 Oftentimes distributions only have one license that only really is applicable to one type of components. 02:03.000 --> 02:10.000 So we also recommend that people use different licenses to cover the different components, the different types of components. 02:10.000 --> 02:19.000 So as you see, there is an example, you have a patchy for instance for your code, cdla for your data and cc by for the document. 02:19.000 --> 02:24.000 We put all of those components into three different classes. 02:24.000 --> 02:28.000 There are corresponding to different use cases. 02:28.000 --> 02:34.000 That's also very important because it's like openness from what point of view to achieve what. 02:34.000 --> 02:46.000 The first class open model, basically addresses the most common use case that people have, which is to be able to use the the model and do whatever they want with it. 02:46.000 --> 02:51.000 That includes fine tuning during performance announcements and so on. 02:51.000 --> 02:59.000 The class two is the open tooling model, gives you a bit more insight as to how the model was trained, so you can do further analysis. 02:59.000 --> 03:07.000 And the class one, which is the last one, is really the open science model, which addresses the use case of reproducibility. 03:07.000 --> 03:16.000 You must be given all the information you need with the right licenses so that you can actually reproduce the same exact model. 03:16.000 --> 03:20.000 We put all of this into a spec, which is available. 03:20.000 --> 03:28.000 So I know it's quite a few cure codes, but so the first one leads you to the morph, which is a specification, which is only about 15 pages. 03:28.000 --> 03:33.000 And then there's the mod, which is a website, there is a registry. 03:33.000 --> 03:42.000 You can use to assess the openness of your model and you can actually record it, make it available on the website for others to find. 03:42.000 --> 03:52.000 It will give you, you basically enter the different components that's in your distribution and what the licenses are for each of your components and it will tell you in which class you fall. 03:53.000 --> 03:58.000 We also have a survey going on in generative AI commands that I invite you to look at. 03:58.000 --> 04:03.000 It will help us, you know, give us information as to what we are going to work on. 04:03.000 --> 04:07.000 This is only one of the products of generative AI commands. 04:07.000 --> 04:11.000 And again, I want to, you know, emphasize that. 04:11.000 --> 04:16.000 The next foundation has one good aspect, which is anybody's welcome to come participate. 04:16.000 --> 04:20.000 There is no, no need to pay anything. You don't have to be a member. 04:20.000 --> 04:28.000 So there is a website dedicated to generative AI commands initially, which is part of the LFAI and data organization. 04:28.000 --> 04:32.000 And if any of this interest you, please join us. 04:32.000 --> 04:33.000 Thank you. 04:33.000 --> 04:35.000 Thank you very much, I know. 04:41.000 --> 04:54.000 That's you. 04:54.000 --> 04:59.000 So now, Martin, it was supposed to be with Paul, but Paul couldn't make it. 04:59.000 --> 05:05.000 So Martin just presented the presentation about Kubernetes AI building themselves. 05:05.000 --> 05:06.000 Thank you very much. 05:06.000 --> 05:08.000 Hello everybody. 05:08.000 --> 05:10.000 Remember now, I'm like a lecturer up here. 05:10.000 --> 05:13.000 If I see you walk and don't, I got a name and shame you. 05:13.000 --> 05:15.000 No, that's only a joke. 05:15.000 --> 05:16.000 All right. 05:16.000 --> 05:25.000 So we, AI folks, businesses are out there and are trying to leverage the advantage with AI. 05:25.000 --> 05:30.000 I wanted to key aspects of that is being able to build better.