coil

Configuration, model, and embedding for COIL (Contextualized Inverted list) type models.

COIL is an information retrieval model that combines the speed of traditional keyword search with the deep understanding of neural networks. It generates context-aware vector representations for every word in a document and stores them in a standard inverted index. During a search, the model only calculates similarity scores for the exact words that appear in both the query and the document.

Originally proposed in COIL: Revisit Exact Lexical Match in Information Retrieval with Contextualized Inverted list.

Classes

CoilConfig([query_length, doc_length, ...])	Configuration class for COIL models.
CoilEmbedding(embeddings[, scoring_mask, ...])	Dataclass containing embeddings and the encoding for COIL models.
CoilModel(config, *args, **kwargs)	Multi-vector COIL model.
CoilOutput([scores, query_embeddings, ...])	Dataclass containing the output of a COIL model.
UniCoilConfig([query_length, doc_length, ...])	Configuration class for UniCOIL models.
UniCoilModel(config, *args, **kwargs)	Single-vector UniCOIL model.