Back to blog

Dev Diary #14 - Public Dataset Explorer & Phi

2024-07-12

Thanks to everyone who has been trying out Glaive and providing feedback. We are contintuing to push fixes but we also have some new things to show you.

First up, we have built a new site called explore.glaive.ai that allows you to browse published datasets and see the kinds of synthetic datasets Glaive can produce. We have also added a new base model called Phi Mini 128k.

explore.glaive.ai

One of the challenges we noticed was that users couldn't explore the types of datasets Glaive can generate without signing up and creating their own.

To address this, we built explore.glaive.ai. This is a new site that allows you to browse published datasets and see the kinds of synthetic datasets Glaive can produce.

You can also fork and import these datasets into your Glaive account to use as a starting point for your own projects.

Phi Base Model

We're also excited to introduce the Phi Mini 128k as a new base model. This is our smallest base model yet and therefore the cheapest and most performant.

Our training pipeline for this model is in beta and it struggles with general purpose tasks, so if you're going to use it, be specific.

Changelog

  • New accounts will have a dataset pre-loaded to help them get started
  • The type of edit that was performed is now shown in the version history
  • Deployed explore.glaive.ai which hosts published dataset versions
  • Fork and import versions from explore.glaive.ai into your account
  • Added Phi Mini 128k as a base model
  • Fixed a bug where variables would be passed even if simple schema editing was enabled
  • Fixed a bug where the back button would not work when trying to leave the site
  • Fixed email branding for login and added a welcome email