Module meta_embed 

MetaEmbed-specific search helpers.

MetaEmbed (ICLR 2026) stores K pre-computed Meta Token vectors per document instead of one vector per token. At query time the Matryoshka ordering of the query Meta Tokens allows a budget parameter to trade recall for latency: budget=k is full accuracy; budget=1 is fastest.

meta_embed_search wires together:

1. Optional PLAID candidate pruning.
2. Budgeted MaxSim scoring over the remaining candidates.
3. Top-k selection.

Functions

meta_embed_search

Search a MultiVectorStore using budgeted MaxSim with optional PLAID candidate pruning.