Skip to main content
Version: 0.6

VectorIndex

  • Wrap a Listener so that outputs are searchable
  • Can optionally take a second Listener for multimodal search
  • Applies to Listener instances containing Model instances which output vectors, arrays or tensors
  • Maybe leveraged in superduper queries with the .like operator

Dependencies

Usage pattern

from superduper import VectorIndex

vi = VectorIndex(
    'my-index',
    indexing_listener=listener_1  # defined earlier calculates searchable vectors
)

# or may use multiple listeners
vi = VectorIndex(
    'my-index',
    indexing_listener=listener_1
    compatible_listener=listener_2 # this listener can have `listener_2.active = False`
)

db.apply(vi)

See also