Auto-Patching: Enhancing Multi-Hop Reasoning in Language Models

TL;DR

Auto-Patch is a new method that dynamically modifies internal states of language models during inference to improve their ability to perform multi-hop reasoning, addressing a key challenge in complex question answering.

Contribution

It introduces Auto-Patch, a novel dynamic hidden state patching technique building on PatchScopes, to enhance multi-hop reasoning in large language models during inference.

Findings

Auto-Patch increases solve rate from 18.45% to 23.63%.

It narrows the gap to Chain-of-Thought prompting.

Dynamic hidden state interventions show promise for complex reasoning.

Abstract

Multi-hop questions still stump large language models (LLMs), which struggle to link information across multiple reasoning steps. We introduce Auto-Patch, a novel method that dynamically patches hidden states during inference to enhance multi-hop reasoning in LLMs. Building on the PatchScopes framework, Auto-Patch selectively modifies internal representations using a learned classifier. Evaluated on the MuSiQue dataset, Auto-Patch improves the solve rate from 18.45\% (baseline) to 23.63~$\pm$~0.7\% (3 runs), narrowing the gap to Chain-of-Thought prompting (27.44\%). Our results highlight the potential of dynamic hidden state interventions for advancing complex reasoning in LLMs.