Hierarchical Token Prepending: Enhancing Information Flow in Decoder-based LLM Embeddings
Xueying Ding · Xingyue Huang · Clark Mingxuan Ju · Liam Collins · Yozen Liu · Leman Akoglu · Neil Shah · Tong Zhao
摘要
Large language models produce powerful text embeddings, but their causal attention mechanism restricts the flow of information from later to earlier tokens, degrading representation quality. While recent methods attempt to solve this by prepending a single summary token, they over-compress inform...