New in September: Increasing flexibility and robustness

The past few months have seen a surge in public adoption of cutting-edge ML research, with models like Stable Diffusion and Whisper setting new benchmarks for model capabilities. It has never been a more exciting time to work in machine learning, which makes MLOps even more pressing. In the race to build products and services around new possibilities in AI, execution and operations make winners. We can’t wait to see what you build.

What we’re reading and writing

• Deploying NLP to the cloud with Truss and Kern AI

• AI and the limits of language in Noema

• Comparing BentoML and Truss for model packaging and deployment

• Deploy your Kaggle models using Truss with notebooks from Jesse

• How we deployed stable diffusion on Baseten

Coming soon: Follow us on Twitter to be the first to know when we publish our hosted Whisper model demo app, which will let you transcribe speech with best-in-class accuracy. 

Flexible interfaces for ML models

What do a binary classifier and an image generation model have in common? This isn’t a trick question: not much. So why should their interfaces be the same? Ok, this one is a trick question: it depends on your use case.

Baseten provides a default interface for all models deployed on the platform, which is set by the predict function in the models/model.py file in the model’s Truss packaging. Having a consistent interface is useful because it means you can interact with different models in a predictable way. However, this default interface is too restrictive for some use cases, so in a recent upgrade to the Baseten Python client, we released a backwards-compatible way for users to set their own model interface.

Read all the technical details in our changelog post.

Robust application development workflows

This month, we shipped a way to back up your applications to GitHub for paid accounts on our Starter and Business plans. Combined with the draft environment, this feature lets you back up the latest version of each application in your Baseten workspace. With this GitHub workflow you can follow best practices for application development and make changes to production apps with confidence.

✕

A screenshot of a GitHub repository full of apps built on Baseten

Plus, we’re always improving the app builder with quality of life features like full workspace resource search. And stay tuned for something we’re really excited about: very soon, you’ll be able to write Python code directly in the front-end of your Baseten apps. Full-stack web development without JavaScript!

Inside Baseten

This month we were thrilled to welcome Sidharth, a software engineer, to Baseten. Next month that could be you! We’re hiring for a lead ML engineer and a lead infrastructure engineer.

We’ll be back with an extra-spooky edition of this newsletter in October.

Thanks all,

The team at Baseten