How to use your favorite Embedder in Cheshire Cat.

In the latest versions of the Cheshire Cat there has been refactoring and many features have been added and removed from the core to make it lightweight and performant.
One of the choices made was to “outsource” the embedders by keeping in the core only those that work through APIs (such as Open AI or Cohere), the Dumb Embedder, and those supported by FastEmbed that despite being downloaded locally use ONNX runtime that makes them very lightweight.

What if I want to use another embedder that I saw on MTEB?

No problem! Run the Cheshire Cat and get ready for the journey.

Go to Settings tab, and choose Embedder configurator.

You will be faced with the new list of embedders available in the core, and let’s say that on MTEB you saw a very promising new model.

MTEB is a massive benchmark for measuring the performance of text embedding models on diverse embedding tasks. It includes 56 datasets across 8 tasks and currently summarizes >2000 results on the leaderboard.

Over the past few days I have heard a lot of good things about the new mixedbread.ai models, and I have become very curious to try them out.
In the leaderboard they stand quite high (seventh place) and can be used through SBERT.

We go to the plugins tab and type embedder in the search bar.

It returns us several plugins to use embedders, there is Jina.ai, Hugging Face, Ollama, … and also the one we need right now, SBERT.
Let’s install it!

Now the plugin is installed and enabled, we just need to configure the template we prefer.
We move to the Hugging Face card and copy the address with the button indicated by the red arrow in the image.

After that we go to the Embedders Configurator, choose SBERT and paste in the model name.

After also configuring the LLM (I chose chatGPT) we can start using Cheshire Cat as usual!

If you can’t find the plugin to use your favorite embedder, you can write an issue on the project’s github repo or try to make your own plugin by taking cues from the ones that are already in the list.
Have fun!