FaissSearchConfig

class lightning_ir.retrieve.faiss.faiss_searcher.FaissSearchConfig(k: int = 10, candidate_k: int = 100, imputation_strategy: 'min' | 'gather' | 'zero' = 'gather', n_probe: int = 1, ef_search: int = 16)[source]

Bases: ApproximateSearchConfig

Configuration class for FAISS-based searchers in the Lightning IR framework.

__init__(k: int = 10, candidate_k: int = 100, imputation_strategy: 'min' | 'gather' | 'zero' = 'gather', n_probe: int = 1, ef_search: int = 16) → None[source]

Initialize the FaissSearchConfig.

Parameters:

  • k (int) – Number of top results to return. Defaults to 10.

  • candidate_k (int) – Number of candidates to retrieve before ranking. Defaults to 100.

  • imputation_strategy (Literal["min", "gather", "zero"]) – Strategy for handling missing scores. Defaults to “gather”.

  • n_probe (int) – Number of probes for the IVF index. Defaults to 1.

  • ef_search (int) – Size of the dynamic list used during search. Defaults to 16.

Methods

__init__([k, candidate_k, ...])

Initialize the FaissSearchConfig.

Attributes

SUPPORTED_MODELS

search_class

alias of FaissSearcher