Retrieval Backward Attention without Additional Training: Enhance Embeddings of Large Language Models via Repetition

TL;DR

This paper introduces a backward attention mechanism to improve embeddings in large language models, significantly enhancing zero-shot performance without additional training, demonstrated on the C-MTEB benchmark.

Contribution

It presents a novel backward attention method that boosts embedding quality in pre-trained models without extra training steps.

Findings

Significant performance improvements on C-MTEB tasks

Effective enhancement of zero-shot learning capabilities

Simple implementation without additional training

Abstract

Language models can be viewed as functions that embed text into Euclidean space, where the quality of the embedding vectors directly determines model performance, training such neural networks involves various uncertainties. This paper focuses on improving the performance of pre-trained language models in zero-shot settings through a simple and easily implementable method. We propose a novel backward attention mechanism to enhance contextual information encoding. Evaluated on the Chinese Massive Text Embedding Benchmark (C-MTEB), our approach achieves significant improvements across multiple tasks, providing valuable insights for advancing zero-shot learning capabilities.