Developer guide: Custom embeddings

Labelbox uses embeddings to power similarity search in Catalog. Use this guide to understand how to view and manage the details of your custom embeddings and the embeddings generated by Labelbox in the app UI.

There are two categories of embeddings in Labelbox:

Precomputed embeddings: These are automatically computed embeddings that Labelbox provides
Custom embeddings: These are embeddings that you can specify and import to Labelbox

To view a list of all supported embeddings, see Similarity search.

Regardless of whether the embeddings are autogenerated by Labelbox or custom (imported), each embedding will have the following fields.

Field	Description
`Name`	Name of the embedding
`ID`	ID of the embedding
`Dimensions`	Refers to the size of the vector space in which words, phrases, or other entities are embedded.
`Custom`	Indicates whether the embedding is a `Precomputed embedding` or a `Custom embedding`

Custom embeddings

Create a custom embedding

Before you can upload custom embeddings to Labelbox, you'll need to do the following:

View the limits page to learn the custom embedding limits per workspace and the maximum dimensions allowed per custom embedding.
Create an API key.

Then, follow these steps to create a custom embedding in the app UI:

Go to Schema > Embeddings.
Select + Create
In the modal, specify a Name and the number of Dimensions for your custom embedding. Then, select Create embedding.

After you create your custom embedding, you can view all of the embedding fields in JSON format by clicking the </> button.

Use this Google colab notebook to upload your custom embeddings.

Delete a custom embedding

Follow these steps to delete a custom embedding from the app UI.

Go to Schema > Embeddings.
Select the embedding you want to delete from the list.
Click the gear icon in the top right corner, and select Delete embedding.
Enter "delete" in the free text box to confirm.