bge-m3

Headlines · 3

1. 2d
   kb replaces default RRF fusion with weighted hybrid retrieval

   The kb project developer says relevant cross-language documents were being retrieved by vector search but then eliminated when BGE-M3 and BM25 results were merged with Reciprocal Rank Fusion. The default hybrid path was changed to weighted_fuse, which normalizes each list and divides by the number of retrieval methods that actually returned a document; relevance thresholds, larger candidate pools, and regression tests were also added. The developer retained RRF for comparisons and explicitly noted that no large-scale benchmark was run.

   掘金 · 人工智能 · 🔥 2
2. 4d
   Developer Reports Recursive Memory in Pure-Python Observer-RAG

   In a 掘金·人工智能 first-person post, a developer describes and open-sources Observer-RAG, a compact local RAG implementation using pure Python, apsw, sqlite-vec, and a local BGE-M3 model. The developer reports that a retrieved chunk containing a preserved citation tag led the model to reference a different memory chunk that was not included in that round’s retrieval set. The “emergent recursive memory” explanation remains an unreplicated personal observation rather than a validated result.

   掘金 · 人工智能 · 🔥 0
3. 6d
   RAG Query Expansion with HyDE and Multi-Query Rewriting

   This tutorial demonstrates how HyDE and multi-query rewriting can expand retrieval entry points for complete queries whose wording differs from knowledge-base documents. Using PostgreSQL 16, pgvector 0.8, bge-m3, and RRF fusion, the author reports that multi-query expansion plus the original query achieved 86% recall@5, compared with 72% without expansion and 80% for HyDE alone, though at higher latency. The evaluation covers 100 queries from the author's device knowledge base and therefore is not a general benchmark.

   掘金 · 人工智能 · 🔥 0

Experience and discussion from the community

Share my bge-m3 experienceAsk about bge-m3