VERA: Validation and Enhancement for Retrieval Augmented systems

TL;DR

VERA is a system that evaluates and refines retrieval and response quality in retrieval-augmented language models, significantly improving accuracy and relevance across various model sizes.

Contribution

Introduces VERA, a novel framework that evaluates and enhances retrieval and generated responses to reduce errors in retrieval-augmented systems.

Findings

Improves accuracy of smaller open-source models

Enhances performance of larger state-of-the-art models

Effectively reduces hallucinations and irrelevant information

Abstract

Large language models (LLMs) exhibit remarkable capabilities but often produce inaccurate responses, as they rely solely on their embedded knowledge. Retrieval-Augmented Generation (RAG) enhances LLMs by incorporating an external information retrieval system, supplying additional context along with the query to mitigate inaccuracies for a particular context. However, accuracy issues still remain, as the model may rely on irrelevant documents or extrapolate incorrectly from its training knowledge. To assess and improve the performance of both the retrieval system and the LLM in a RAG framework, we propose \textbf{VERA} (\textbf{V}alidation and \textbf{E}nhancement for \textbf{R}etrieval \textbf{A}ugmented systems), a system designed to: 1) Evaluate and enhance the retrieved context before response generation, and 2) Evaluate and refine the LLM-generated response to ensure precision and minimize errors. VERA employs an evaluator-cum-enhancer LLM that first checks if external retrieval is necessary, evaluates the relevance and redundancy of the retrieved context, and refines it to eliminate non-essential information. Post-response generation, VERA splits the response into atomic statements, assesses their relevance to the query, and ensures adherence to the context. Our experiments demonstrate VERA's remarkable efficacy not only in improving the performance of smaller open-source models, but also larger state-of-the art models. These enhancements underscore VERA's potential to produce accurate and relevant responses, advancing the state-of-the-art in retrieval-augmented language modeling. VERA's robust methodology, combining multiple evaluation and refinement steps, effectively mitigates hallucinations and improves retrieval and response processes, making it a valuable tool for applications demanding high accuracy and reliability in information generation. .