Prolonged Reasoning Is Not All You Need: Certainty-Based Adaptive Routing for Efficient LLM/MLLM Reasoning

TL;DR

This paper introduces Certainty-based Adaptive Reasoning (CAR), a framework that dynamically switches between short and long reasoning in LLMs/MLLMs based on confidence, improving efficiency and accuracy over fixed reasoning strategies.

Contribution

The paper proposes a novel adaptive reasoning framework that uses model perplexity to decide when to perform detailed reasoning, addressing the limitations of prolonged reasoning.

Findings

CAR outperforms fixed reasoning strategies on multimodal and text reasoning tasks.

Using perplexity to trigger reasoning improves efficiency without sacrificing accuracy.

Prolonged reasoning can degrade performance on simpler tasks.

Abstract

Recent advancements in reasoning have significantly enhanced the capabilities of Large Language Models (LLMs) and Multimodal Large Language Models (MLLMs) across diverse tasks. However, excessive reliance on chain-of-thought (CoT) reasoning can impair model performance and brings unnecessarily lengthened outputs, reducing efficiency. Our work reveals that prolonged reasoning does not universally improve accuracy and even degrade performance on simpler tasks. To address this, we propose Certainty-based Adaptive Reasoning (CAR), a novel framework that dynamically switches between short answers and long-form reasoning based on the model perplexity. CAR first generates a short answer and evaluates its perplexity, triggering reasoning only when the model exhibits low confidence (i.e., high perplexity). Experiments across diverse multimodal VQA/KIE benchmarks and text reasoning datasets show that CAR outperforms both short-answer and long-form reasoning approaches, striking an optimal balance between accuracy and efficiency.