
๐ช๐ต๐ฎ๐ ๐๐ฟ๐ฒ ๐๐ถ-๐ฒ๐ป๐ฐ๐ผ๐ฑ๐ฒ๐ฟ๐ ๐ฎ๐ป๐ฑ ๐๐ฟ๐ผ๐๐-๐ฒ๐ป๐ฐ๐ผ๐ฑ๐ฒ๐ฟ๐? ๐น Bi-encoders: โ Two separate neural networks โ One encodes your query โ Another encodes documents โ Compare the embeddings using similarity metrics ๐น Cross-encoders: โ Single neural network โ Takes query + document together as input โ Outputs a relevance score directly โ More accurate but slower Now hereโs how they work in RAG: ๐ธ ๐๐ถ-๐ฒ๐ป๐ฐ๐ผ๐ฑ๐ฒ๐ฟ๐ (Stage 1: Fast Retrieval) 1. Your query: โHow to reduce RAG costs?โโฆ
No discussion yet. Be the first to share your thoughts!