Chain-of-Thought Augmentation with Logit Contrast for Enhanced Reasoning in Language Models

TL;DR

This paper introduces logit contrast techniques to augment chain-of-thought prompting, aiming to improve reasoning and compositional generalization in language models, though further validation across datasets is needed.

Contribution

It proposes input-based contrasting methods inspired by context-aware decoding to enhance reasoning capabilities in language models using chain-of-thought prompting.

Findings

Initial improvements in reasoning performance.

Potential for better compositional generalization.

Further work needed for stability across datasets.

Abstract

Rapidly increasing model scales coupled with steering methods such as chain-of-thought prompting have led to drastic improvements in language model reasoning. At the same time, models struggle with compositional generalization and are far from human performance on many reasoning-based benchmarks. Leveraging the success of chain-of-thought prompting, and also taking inspiration from context-aware decoding (CAD), we explore input-based contrasting methods to further encourage the type of reasoning induced by chain-of-thought prompting. While work remains to stabilize these results across datasets and models, the improvements we find warrant further investigation into input-based steering methods for context-aware reasoning.