Changelog #034
October 07, 2023Follow along on Twitter for detailed updates and improvements made to Banana.
Automation API
Tired of clicking buttons in a UI? Want to step up your GPU game?
We're proudly launching our Automation API, your way to integrate with Banana to automatically deploy and manage your AI apps. Currently supported features:
- GET projects -> list your projects
- GET project -> get a project by ID
- PUT project -> update a project's minReplicas, maxReplicas, idleTimeout, and inferenceTimeout
Monorepo support
Users on Banana now no longer need to have a unique repo for each project.
You may now use a glorious monorepo!
Simply point to your project root and Dockerfile path to your potassium app within your repo and we'll build from there.
Feature Calculator
There are a lot of levers to pull when serving GPUs, and we want to be more helpful in helping you choose if Banana gives you those levers.
We've created Banana And You, a calculator for thinking through if we're the right tool for you.
"Models" are now called "Projects"
We've renamed the "Model" construct in Banana to "Project"
Here's why:
- Most users run additional logic on our servers beyond just single-model inference, so the word model didn't feel right.
- In practice, AI backends often contain multiple models
- We have visions of Projects on Banana being multi-model, multi-endpoint, multi-device distributed deployments of your AI backend. All in a single project and served via Potassium.
This change is minor, including the deprecation of Model Keys, since deployments already have unique endpoints and API keys for authentication. These changes are reflected in the Banana SDKs, with both Python and Node dropping the model key argument in recent breaking-change release 6.0.0
--
If you have any feature suggestions, improvements, or bug reports, send us a message or let us know in #support or #feature-requests on Discord.