In the recent GigaOm Sonar Report on Vector Databases where Vespa came out as Leading, one of the criteria where we scored Exceptional was Embedding Flexibility.

What’s so great about the embedding flexibility in Vespa? You have the choice of doing embeddings in four ways:

• On your own, outside Vespa: Just pass tensors directly in documents and queries.
• With your own model, run by Vespa: Add the model to the application package and reference it in embed expressions.
• With a model provided by Vespa: Reference a model on the model hub in your embed expression.
• With custom code doing what you want in a custom Embedder, or, if you want full control over the process, a custom Searcher and Docproc

In addition to creating single embeddings for a field, all these methods also allow you to create and index a collection of embeddings for a single document, either by using an embedding model that creates an embedding for each token of the text, by embedding an array of strings, or doing both at the same time. You can use multiple of these methods at the same time for different fields, and change method at any time without changing any other aspect of the application. This lets you get started easily with embeddings while also empowering you to add more sophisticated methods gradually.

And if you add fields that derive an embedding for a field to an existing application, they will be automatically populated. Combined with the support for having multiple embeddings in the same document and choosing which ones to use in queries and ranking, this makes it exceptionally easy to experiment with new embedding models.