Hi, Why dot_product over cosine?

Dot product is computationally faster for unit vectors since cosine similarity of unit vectors equals their dot product, but Elasticsearch can optimize the calculation. For unit vectors: cosine(A,B) = dot(A,B) since ||A|| = ||B|| = 1.

Updated Elasticsearch Mapping (v8+)

{
  "mappings": {
    "properties": {
      "vector_field": {
        "type": "dense_vector",
        "dims": 384,  // your vector dimensions
        "similarity": "dot_product"
      }
    }
  }
}