Superposition Prompting: Improving and Accelerating Retrieval-Augmented Generation

TL;DR

Superposition prompting is a novel method that enables large language models to process multiple input paths in parallel, reducing inference costs and improving accuracy in retrieval-augmented generation without fine-tuning.

Contribution

It introduces superposition prompting, a new RAG technique that enhances efficiency and accuracy of pre-trained LLMs by parallel processing of prompt paths without fine-tuning.

Findings

93x reduction in compute time on NaturalQuestions-Open dataset

43% accuracy improvement over naive RAG with MPT-7B

Effective across various question-answering benchmarks

Abstract

Despite the successes of large language models (LLMs), they exhibit significant drawbacks, particularly when processing long contexts. Their inference cost scales quadratically with respect to sequence length, making it expensive for deployment in some real-world text processing applications, such as retrieval-augmented generation (RAG). Additionally, LLMs also exhibit the "distraction phenomenon", where irrelevant context in the prompt degrades output quality. To address these drawbacks, we propose a novel RAG prompting methodology, *superposition prompting*, which can be directly applied to pre-trained transformer-based LLMs *without the need for fine-tuning*. At a high level, superposition prompting allows the LLM to process input documents in parallel *prompt paths*, discarding paths once they are deemed irrelevant. We demonstrate the capability of our method to simultaneously enhance time efficiency across a variety of question-answering benchmarks using multiple pre-trained LLMs. Furthermore, our technique significantly improves accuracy when the retrieved context is large relative the context the model was trained on. For example, our approach facilitates a 93x reduction in compute time while *improving* accuracy by 43% on the NaturalQuestions-Open dataset with the MPT-7B instruction-tuned model over naive RAG.