Your Fine-Tuned Model Is Your IP: Open Model Strategy for Developers

— Hello friends. Thank you for joining us today. My name is Yina Arenas and I am the corporate vice president for Microsoft boundary and I have the honor and pleasure to have Ling Zhao with us today, CEO of Firework. Thank you Ling for joining us. — Thank you for having me. — It's been fantastic to see our partnership over the last couple of months, but I would love for us to start our conversation today just talking about open source and what you're seeing in Firework. — We are seeing explosive usage and interest on open source from multiple different angles. First of all, um, the application space and agentic space has been growing really fast and everyone want to have control of what is their real value, a long-term durable value. And part of that control lies in their private data. There's so much private data locked inside application, locked inside enterprise and the way to turn um, those private data into intelligence unique to uh, to them is to fine tune the model and a customized model. And after a customized model, they need to bring into production and scale quickly is absolutely the top concern. So, so that's where we come into the picture to help those customized model run the fastest uh, with the highest quality, with the highest cost efficiency. So, app developers, agentic developers, they can re-quickly reach billions of consumers. — A couple of years back when you were talking about fine tune models, people were like not necessarily thinking it was worth the investment because the soda models were advancing so fast that they by the time that they finished fine tuning, the new model was already better. But like the point that you're bringing here is so important. It is how you customize a model to work for your scenario, right? Your data, your tools, knowing of your environment, all of that information being able to put it together in a model so that it works, whether it is on agentic scenarios or elsewhere in AI-powered applications, but that it knows about you, right? And your IP. I think actually that's one area where we're going to see a lot of movement in the market in the coming years, where every company is going to become a model company, because they're going to be producing their own models with the knowledge that they can only have. That's their differentiation. — Absolutely. I totally agree with that because we have been powering so many startups and generative companies built on top of us. I think one of the pattern design pattern we see is they don't think about model just as a black box API or commodity. They really think about model as a critical part of their product and then they're doing product model co-design. The fundamental reason is the data distribution of the training data goes into model and the data distribution of the data from the application side by nature are misaligned because those are designed by two different companies. And that misalignment is going to leave a lot of quality, speed, and cost optimization on the table. As you mentioned, model is your IP. Once you customize it, it continues grow with the product, evolve with the product. That is IP. So we absolutely see a lot of adoption of product philosophical design along with the model in this current wave. — And that's one of the strengths of our partnership. Because like if you look at the process that our customers need to do today is like they need a tool for creating the maybe the evaluations, a tool for the inferencing, a tool for doing the fine-tuning. Then like you end up stitching multiple vendors, multiple contracts, and then like by the time that you're done, like it's an amalgamation of processes in order to get an output, right? And now with Fireworks on Microsoft Foundry, you get the best of both worlds. You get the super optimized inference that Fireworks has, all of the best open source models, the ability to fine-tune them, bring those optimized weights into Foundry, do the inferencing, and do the complete loop of like not just having the model, but all like and the inferencing, but like connecting that to an agentic solution, connecting that to tools, to memory, to knowledge, taking that into production, then like doing the governance of that application once it is running in prod, continuously improving that based on the evaluations. That's super powerful today. And a huge shift from like having to stitch multiple vendors together to just like leverage what we have to offer together. — That's where it laid a very solid foundation of our partnership because our value fully aligned. I see our shared value as we want to optimize for the developers. their end-to-end journey. And at Fireworks, we have been laser focused on, you know, optimizing for this customized model. And there's a much wider

ecosystem for the end-to-end journey, and it is a solving a fundamental problem have one place to go all the way from ideation of experimentation to post-product market fit scaling. That's a long journey. And that's where we see Microsoft Foundry is providing us such a comprehensive toolkit to the developer where we can plug in to offer the best customized model inference. — I mean, in your platform, you are sharing this earlier with me today that there is like trillions and trillions of tokens flowing through, maybe more than just trillions and trillions. — 30 trillion tokens today. — like mind-blowing, right? And we have in our partnership lots of customers that started using our product before GA and it's just really great to see how they are now reaching scale and going from like customizing a model, bringing it into the inferencing and now getting to big token inferencing. — It started from our deep conviction of believing the future that the industry has to activate every single company's private data and bring that to build their proprietary intelligence and they use that also to make the model super fast and cost efficient to scale. We start there, but after we announced our preview, it is just taking off. And the many customers are actually banging our doors and asking when they can get on board. I think this is just the beginning of our journey into growing to a huge business together. — Absolutely. I mean and the customer value is so huge, right? As a developer, as an architect, as a business decision, like now taking your data, your tools, your processes and being able to fine tune models to like bake that in into the model, make them super optimized for your scenario. It's not about hitting the industry benchmarks, right? It's about a model that knows your business and then being able to leverage all of the optimized inference that Fireworks has with all of the platform components that Foundry offers around the reliability, about the evaluations, observability, the taking that into production and being able to continuously improve. It's just like fantastic to see. Like the customers I hear saying wanting to come in, wanting to get into the product and seeing their scenarios shine through this partnership. — Absolutely, yes. — It was amazing to see customers getting early on into the preview and already doing production level workloads. A long tail of customers, as you're saying, banging our door to like get into the preview. And now, which is the exciting thing that we're here together to announce, is the general availability of Fireworks in Microsoft Foundry, and how customers can not only take all of the greatness that Fireworks. ai offer, but also Foundry has to offer in a simplified way, one endpoint, one contract, and one journey, one platform. So, I'm super excited. Thank you, Lin. It's been an amazing partnership with your team. — I firmly believe a durable partnership comes with shared value and shared vision. And this is I see what unique about us working together. And there's only 1% journey is done. And there's a lot ahead of us to grow into. — Absolutely. So, if you're a developer, if you're a data scientist, the world has shifted from not thinking too much about fine-tuning a couple of years ago to like fine-tuning is what is going to be your IP, your mode, because it understands your data, it understands your scenarios. And with the partnership between Fireworks and Microsoft Foundry, we have the best of both worlds to offer, where you can customize the model, deploy it, and run it at scale, observe it, continuously improve it, and you know, power your applications from idea all the way to production. Friends, thank you for joining us today. Thank you, Lin, for an amazing conversation. — Thanks, Lin. — And everyone, if you're at Build, don't miss — catching up with our sessions. I can't wait to see what you build with Fireworks and Microsoft Foundry. —