The shortest explanation on how to build embeddings for queries and documents at scale with Vespa ! 😍

1. Download and convert to onnx any Hugging Face sentence transformer model(s)

2. Declare the model(s) with an id in services.xml

=> id=”bert”

3. Use the model(s) id to declare tensor field(s) built from any text field(s)

=> indexing: input myTextField | embed bert

4. Deploy the application package on docker or Vespa cloud

5. Use the model(s) id to embed the query text

=> input.query(myEmbedding)=embed(bert, “Hello world”)

6. Use ranking(s) to build mixed BM25/ANN expressions