Pass IndexThe State of AISign in

Chunking

Cutting documents into pieces small enough to embed and retrieve usefully.

how it works · the vocabulary

splitting

The least glamorous and most consequential decision in a retrieval system. Cut too small and an answer loses its context; too large and the search returns mostly irrelevance. Most retrieval failures blamed on the model are chunking failures.

Nearby

AgentAgent memoryAlignmentAutoregressiveBM25Cold startComputer useContext window