Documentation
¶
Overview ¶
Package reranking provides cross-encoder reranking for search results.
Package reranking provides cross-encoder reranking for search results.
Package reranking provides cross-encoder reranking for search results. Uses MS-MARCO MiniLM L6 v2 cross-encoder model for relevance scoring.
Package reranking provides cross-encoder reranking for search results.
Index ¶
- Constants
- Variables
- type APIConfig
- type APIService
- func (s *APIService) Close() error
- func (s *APIService) Rerank(query string, candidates []Candidate, limit int) ([]RerankResult, error)
- func (s *APIService) RerankByScore(query string, candidates []Candidate, limit int) ([]RerankResult, error)
- func (s *APIService) Score(query, document string) (rawScore, normalizedScore float64, err error)
- type Candidate
- type Config
- type RerankResult
- type Reranker
- type Service
- func (s *Service) Close() error
- func (s *Service) Rerank(query string, candidates []Candidate, limit int) ([]RerankResult, error)
- func (s *Service) RerankByScore(query string, candidates []Candidate, limit int) ([]RerankResult, error)
- func (s *Service) Score(query, document string) (rawScore, normalizedScore float64, err error)
Constants ¶
const ( // ModelName is the human-readable name for the cross-encoder model ModelName = "ms-marco-MiniLM-L6-v2" // ModelVersion is the short version string for identification ModelVersion = "msmarco-v2" // MaxSequenceLength is the maximum combined query+document token length MaxSequenceLength = 512 )
const ( // DefaultCandidateLimit is the default number of candidates to rerank. DefaultCandidateLimit = 100 // DefaultResultLimit is the default number of results to return after reranking. DefaultResultLimit = 10 )
Variables ¶
var NewONNXReranker func(alpha float64) (Reranker, error)
NewONNXReranker creates an ONNX-based reranker. On Windows, this always returns nil (ONNX runtime unavailable). On other platforms, it creates the cross-encoder service.
Functions ¶
This section is empty.
Types ¶
type APIConfig ¶
type APIConfig struct {
BaseURL string
APIKey string
Model string
Alpha float64
Timeout time.Duration
}
APIConfig holds configuration for the API-based reranker.
func DefaultAPIConfig ¶
func DefaultAPIConfig() APIConfig
DefaultAPIConfig returns sensible defaults for the API reranker.
type APIService ¶
type APIService struct {
// contains filtered or unexported fields
}
APIService provides cross-encoder reranking via an external API (Cohere-compatible).
func NewAPIService ¶
func NewAPIService(cfg APIConfig) (*APIService, error)
NewAPIService creates a new API-based reranker.
func (*APIService) Close ¶
func (s *APIService) Close() error
Close releases resources. API service has no persistent resources to release.
func (*APIService) Rerank ¶
func (s *APIService) Rerank(query string, candidates []Candidate, limit int) ([]RerankResult, error)
Rerank reranks candidates using combined bi-encoder + cross-encoder scores.
func (*APIService) RerankByScore ¶
func (s *APIService) RerankByScore(query string, candidates []Candidate, limit int) ([]RerankResult, error)
RerankByScore reranks candidates sorted by pure cross-encoder score only.
type Candidate ¶
type Candidate struct {
Metadata map[string]any
RerankInfo map[string]float64
ID string
Content string
Score float64
}
Candidate represents a search result candidate for reranking.
type Config ¶
type Config struct {
// Alpha is the weight for combining scores (0.0-1.0)
// Higher values favor cross-encoder scores, lower values favor bi-encoder scores
Alpha float64
}
Config holds configuration for the reranking service.
func DefaultConfig ¶
func DefaultConfig() Config
DefaultConfig returns sensible defaults for reranking.
type RerankResult ¶
type RerankResult struct {
Metadata map[string]any
ID string
Content string
OriginalScore float64
RerankScore float64
CombinedScore float64
OriginalRank int
RerankRank int
RankImprovement int
}
RerankResult represents a reranked search result.
type Reranker ¶
type Reranker interface {
// Rerank reranks candidates using combined bi-encoder + cross-encoder scores.
// Returns up to limit results sorted by combined score.
Rerank(query string, candidates []Candidate, limit int) ([]RerankResult, error)
// RerankByScore reranks candidates sorted by pure cross-encoder score only.
RerankByScore(query string, candidates []Candidate, limit int) ([]RerankResult, error)
// Score scores a single query-document pair.
// Returns raw cross-encoder logit and normalized (0-1) score.
Score(query, document string) (rawScore, normalizedScore float64, err error)
// Close releases model resources.
Close() error
}
Reranker defines the interface for cross-encoder reranking implementations. Implementations include the ONNX-based local Service and the API-based APIService.
type Service ¶
type Service struct {
// Weight for combining scores: combined = alpha*rerank + (1-alpha)*original
// Default 0.7 favors cross-encoder score
Alpha float64
// contains filtered or unexported fields
}
Service provides cross-encoder reranking functionality.
func NewService ¶
NewService creates a new cross-encoder reranking service. Note: ONNX runtime must be initialized before calling this (via embedding.NewService).
func (*Service) Rerank ¶
Rerank reranks candidates using the cross-encoder model. Takes a query and list of candidates, returns reranked results.
func (*Service) RerankByScore ¶
func (s *Service) RerankByScore(query string, candidates []Candidate, limit int) ([]RerankResult, error)
RerankByScore reranks candidates and returns sorted by pure cross-encoder score. Useful when you want to completely replace bi-encoder ranking.
Source Files