MARAGS: A Multi-Adapter System for Multi-Task Retrieval Augmented Generation Question Answering

TL;DR

MARAGS is a multi-adapter retrieval augmented generation system designed for complex question answering tasks, effectively integrating diverse data sources and adapters to improve performance on multi-task RAG challenges.

Contribution

The paper introduces MARAGS, a novel multi-adapter system that enhances multi-task retrieval augmented generation for question answering, demonstrating competitive results in a major competition.

Findings

Achieved 2nd place in Task 1 of CRAG competition.

Achieved 3rd place in Task 2 of CRAG competition.

Utilizes multiple adapters and a cross-encoder for effective passage ranking.

Abstract

In this paper we present a multi-adapter retrieval augmented generation system (MARAGS) for Meta's Comprehensive RAG (CRAG) competition for KDD CUP 2024. CRAG is a question answering dataset contains 3 different subtasks aimed at realistic question and answering RAG related tasks, with a diverse set of question topics, question types, time dynamic answers, and questions featuring entities of varying popularity. Our system follows a standard setup for web based RAG, which uses processed web pages to provide context for an LLM to produce generations, while also querying API endpoints for additional information. MARAGS also utilizes multiple different adapters to solve the various requirements for these tasks with a standard cross-encoder model for ranking candidate passages relevant for answering the question. Our system achieved 2nd place for Task 1 as well as 3rd place on Task 2.