Loopwise contrastive decoding improves reasoning in looped language models
LoopCD: Loop-wise Contrastive Decoding for Improving Reasoning in Looped Language Models
Computation and Language
Summary
Looped language models use repeated internal steps to improve their thinking, but sometimes they get confused on tricky parts. The authors found a way to fix these confusing spots by comparing earlier and later guesses during the model’s process at the time of answering questions. This technique makes the model’s reasoning clearer and more reliable without needing extra training or slowing it down much. Testing showed it helps these models perform better on different reasoning tasks.
What this means in practice
- •For ai system developers: Improve the reasoning accuracy of looped language models in AI applications by integrating loop-wise contrastive decoding during inference.
- •For software engineers: Enhance natural language processing tools that rely on recursive refinement by reducing errors on hard reasoning tokens without retraining models.
Authors
Byeongho Yu, Junhyuk So, Eunhyeok Park
Abstract
Looped Language Models (LoopLMs) perform "latent reasoning" by recursively refining internal latent representations with shared weights, offering a more effective alternative to explicit verbal reasoning. Despite their effectiveness, we find that LoopLMs remain prone to loop instability: unstable refinement across iterations can produce localized uncertain "hard" tokens associated with reasoning errors. To address this, we propose LoopCD, loop-wise contrastive decoding that enhances the reasoning performance of LoopLMs by intervening on these tokens at inference time. Specifically, we exploit the internal dynamics of LoopLMs and contrast the logits from earlier iterations with logits from the last refined iteration to form the final sampling distribution. We find that this strategy is highly efficient, introducing only negligible inference overhead and requiring no additional training, while effectively improving reasoning performance by naturally refining reasoning-critical hard tokens. Extensive experiments show that our method improves the performance of recent representative LoopLMs across various reasoning tasks.