A retrieval model is a system, often built on a Transformer architecture, designed to search through and fetch relevant information from a large external data source, such as a document collection or database, rather than generating an answer purely from what it memorized during training. Given a query, it returns the passages or records most likely to be relevant to answering it.
Retrieval models are commonly paired with large language models in a retrieval-augmented generation (RAG) setup, where the retrieved information is fed into the language model's context before it generates a response. This combination lets the language model ground its answers in factual, up-to-date, external information, rather than relying solely on its own training data, which meaningfully reduces the risk of hallucination, especially on questions involving specific or recent facts the model was never trained on.
