Mmr

@Serializable

data class Mmr(val diversity: Float? = null, val candidatesLimit: Int? = null)

Maximal Marginal Relevance reranking, applied after a nearest search using the same query vector.

It trades relevance for variety: without it, ten results about the same thing all rank highly, which is usually not what someone searching wanted to read.

Constructors

Mmr

constructor(diversity: Float? = null, candidatesLimit: Int? = null)

Properties

candidatesLimit

@SerialName(value = "candidates_limit")

val candidatesLimit: Int?

how many candidates to rerank; the query's limit when null.

diversity

@SerialName(value = "diversity")

val diversity: Float?

0 favours relevance, 1 favours dissimilarity between results; Qdrant defaults to 0.5.