Announcing Blueprint: fine-tuning and serving infrastructure for developers

Baseten makes it easier to solve problems with machine learning. We started by streamlining ML deployment so companies could serve models at low cost and quick pace. Feedback has been spectacular: startups like Patreon, Motive, and Laurel reduced resource-draining AI tasks by hundreds of engineering hours and hundreds of thousands of dollars. With more time and cash on hand, they can focus on delivering great experiences for customers.

The best part: the pool of creators is expanding with this year’s explosion in the quality and amount of open source foundation models available to everyone. You no longer have to be a VC-backed founder to make a dent in the ML universe; every tinkerer with a laptop can now build applications on foundation models.

But there’s still one problem. It’s hard to customize these models without prior experience in building, training, and serving. If we made it even easier for anyone to deploy open-source models, the amount of innovation, value, and earth-shattering creativity in the ML world would skyrocket.

That’s why we’re launching Blueprint, a model fine-tuning and serving platform. With Blueprint’s API, you can extend open source models by adding and training on your own datasets, starting with Dreambooth and full Stable Diffusion. When you’re done fine-tuning, you can deploy models within Baseten’s serverless GPU platform—no custom infrastructure needed. You can also use Blueprint to build your own APIs using serverless functions developed in the platform’s web IDE.

Blueprint is ready to use today, and every new user gets 4 hours of credits to get started for free. Sign up today to get started!

Fine-tuned models on Blueprint

We can’t wait to see what you build with Blueprint. We’ve made a few resources to help you get started:

We hope you give Blueprint a try! And join our Discord community to talk to the team that built Blueprint: We want to hear if you get stuck, have questions, or have other feedback.