Method Rerank

Namespace

LMKit.Embeddings

Assembly

LM-Kit.NET.dll

Rerank(string, PartitionSimilarity, float, CancellationToken)

Reranks the similarity score of a single PartitionSimilarity synchronously by invoking the reranking model and updating its RerankScore property with the new blended score.

public void Rerank(string query, PartitionSimilarity partitionSimilarity, float rerankAlpha = 0.5, CancellationToken cancellationToken = default)

Parameters

query string

The input text query used for reranking. Must not be null.

partitionSimilarity PartitionSimilarity

The partition similarity object whose RerankScore property will be updated. Must not be null.

rerankAlpha float

A blending factor between the original similarity (top[i]) and the reranker model score (rerank[i]), in the range [0 - 1]. Computed as:

final_score = (α × top[i]) + ((1 − α) × rerank[i])

For example, with α = 0.5:

final_score = (0.5 × 0.95) + (0.5 × 0.07) = 0.475 + 0.035 = 0.510

A typical default is 0.5 for equal weighting.

cancellationToken CancellationToken

A token to cancel the operation if needed.

Exceptions

ArgumentNullException

Thrown if query or partitionSimilarity is null.

Rerank(string, IEnumerable<PartitionSimilarity>, float, CancellationToken)

Reranks the similarity scores of multiple PartitionSimilarity instances synchronously by invoking the reranking model and updating each instance’s RerankScore property with its new blended score.

public void Rerank(string query, IEnumerable<PartitionSimilarity> partitionSimilarities, float rerankAlpha = 0.5, CancellationToken cancellationToken = default)

Parameters

query string

The input text query used for reranking. Must not be null.

partitionSimilarities IEnumerable<PartitionSimilarity>

A collection of partition similarity objects whose RerankScore properties will be updated. Must not be null or empty.

rerankAlpha float

A blending factor between the original similarity (top[i]) and the reranker model score (rerank[i]), in the range [0 - 1]. Computed as:

final_score = (α × top[i]) + ((1 − α) × rerank[i])

For example, with α = 0.5:

final_score = (0.5 × 0.95) + (0.5 × 0.07) = 0.475 + 0.035 = 0.510

A typical default is 0.5 for equal weighting.

cancellationToken CancellationToken

A token to cancel the operation if needed.

Exceptions

ArgumentNullException

Thrown if query is null or partitionSimilarities is null or empty.