ConceptMatcher¶

class linear_relational.ConceptMatcher(model, tokenizer, concepts, layer_matcher=None, map_activations_fn=None)[source]¶

Match concepts against subject activations in a model

concepts¶

layer_matcher¶

layer_name_to_num¶

map_activations_fn¶

model¶

query(query, subject)[source]¶

query_bulk(queries, batch_size=4, verbose=False)[source]¶

tokenizer¶

class linear_relational.ConceptMatchQuery(text: str, subject: str | int | Callable[[str, list[int]], int])[source]¶

subject¶

text¶

class linear_relational.QueryResult(concept_results: dict[str, linear_relational.ConceptMatcher.ConceptMatchResult])[source]¶

property best_match¶

concept_results¶

class linear_relational.ConceptMatchResult(concept: str, score: float)[source]¶

concept¶

score¶