Retrieval quality deep-dive: NDCG, MRR, HitRate, when the retriever fails
Retrieval quality deep-dive: NDCG, MRR, HitRate, when the retriever fails The retriever returns the…
Re-ranking in RAG Pipelines: Cross-Encoders and Two-Stage Retrieval
Two-stage retrieval for perfect relevance: bi-encoder for fast retrieval, cross-encoder for intelligent re-ranking of top-k. Quality metrics (nDCG, MRR, recall@k), latency-quality trade-offs, and real-world implementation with sentence-transformers + RediSearch on local Ollama.
What do you think about this article?
At a Glance
Two-stage retrieval for perfect relevance: bi-encoder for fast retrieval, cross-encoder for intelligent re-ranking of top-k. Quality metrics (nDCG, MRR, recall@k), latency-quality trade-offs, and real-world implementation with sentence-transformers + RediSearch on local Ollama.
Discuss this article in the community
Join the Telegram group to discuss with other developers, ask questions and share your experiences.
Did you like this article?
Explore other content on the blog or discover my projects
Commenti
Caricamento commenti...
Sign in to comment