Reducing Representational Interference in Retrieval Embeddings with Trainable Prefix Tokens
arXiv:2611.07298v1 Announce Type: new Abstract: Embedding models compress variable-length text into fixed-size vector representations, but this compression can …