Welcome to the inaugural Glaive Dev Diary! We are aiming to put these out weekly to give you a sense of what we accompished in the past week and what is coming up next.

What is Glaive?

To put it simply we are trying to make it really, really easy for anyone to build, measure, and improve their own models. Fundamentally, we are betting on the idea of many small but focused models being more effective than a single large model. The nice thing about this approach is that its cheaper, faster, and you stay in control of your data.

If that sounds like something you would be interested in, then maybe we can help you out.

Private Beta

For the past couple of months we have been working on Glaive in private and its about time we get it in front of more people. So we are going to start letting more teams in to test out the platform and give us feedback.

You can join the waitlist or send us an email explaining how you want to use the platform.

Changelog

Knowledge Graphs Beta

During the model creation process you will now generate a knowledge graph of the topics your model will be trained on. This is our attempt to help you visualize the "knowledge" your model will "learn". Right now you can only view the graph that is generated, but soon you will be able to manually edit the graph to add or remove topics.

Docs and API

Considering there is a lot of new stuff to learn and it can be tricky to learn what buttons to press and levers to pull, we have started to put together some documentation. While rather incomplete right now, it does have a nice tutorial on how to start from nothing and get to your own model running locally with Glaive. You can find the docs here.

We also put out an OpenAPI spec for the Glaive API. While you can still do most things through the web interface, the API is required for certain features like:

• Downloading models and datasets programatically
• Uploading model outputs and feedback so you can track your model's performance

Fixes

• Deletes actually work now
• Website is now more mobile friendly
• Previews can now better handle JSON objects (s/o Erik)

Known Issues

• Switching teams causes an annoying flicker
• Knowledge graphs sometimes have overlapping nodes