BCEmbedding
A bilingual and crosslingual embedding and reranker model pair for retrieval-augmented generation in English and Chinese, developed by NetEase Youdao and used in its own RAG products.
BCEmbedding is a pair of models for retrieval-augmented generation: an EmbeddingModel that generates semantic vectors for search and question answering, and a RerankerModel that refines search results and ranking. Both are bilingual and crosslingual across English and Chinese, and the RerankerModel additionally supports Japanese and Korean and can rerank long passages beyond 512 tokens.
It is the foundation of Youdao's own RAG implementation, QAnything, which is used in products such as Youdao Speed Reading and Youdao Translation, and reports strong results on semantic representation benchmarks in MTEB and on RAG evaluations in LlamaIndex. It targets common business scenarios such as education, medical, law, finance, literature, FAQ, textbook and general conversation use, without requiring fine-tuning.
It integrates with LangChain and LlamaIndex, and is written in Python under the Apache 2.0 license.