HyperThink: Text-to-Parameter Hypernetworks for Efficient Reasoning

2026-10-02

AI Summary

This study addresses the high latency and computational overhead associated with long chain-of-thought (CoT) reasoning in large language models by proposing a text-to-parameter reasoning amortization mechanism. Methodologically, a lightweight hypernetwork is employed to compress multi-step reasoning computations into a single parameter update, thereby eliminating CoT generation during inference. Furthermore, a vector-quantized decoder is integrated to enhance robustness, while end-to-end self-training facilitates efficient optimization. Experimental results demonstrate that the proposed approach substantially reduces token consumption across mathematical and general reasoning tasks. Notably, it achieves a significantly improved trade-off between reasoning accuracy and inference speed in near-CoT-free scenarios.

Abstract

Long-form thinking traces can substantially improve the multi-step reasoning performance of large language models (LLMs), but they introduce high inference-time overhead, with latency dominated by sequential decoding. We propose HyperThink, a text-to-parameter approach that amortizes this reasoning computation into a single query-conditioned parameter update: a lightweight hypernetwork reads the question and predicts updates to a small subset of the base LLM's parameters, while a vector-quantized decoder constrains them to a finite set of reusable patterns to improve robustness and transfer. Trained end-to-end on outputs from the base model itself, HyperThink eliminates long thinking traces at test time: after one hypernetwork forward pass, the adapted model generates a concise step-by-step solution and final answer without an intermediate trace, using far fewer tokens while retaining strong reasoning performance. Empirically, HyperThink improves the low-latency region of the accuracy-latency trade-off on mathematical and general reasoning tasks, with its strongest gains in the near-non-thinking regime.

Problem

• multi-step reasoning
• inference latency
• large language models
• thinking traces
• efficiency

Innovation

• Hypernetworks
• Text-to-Parameter
• Vector Quantization
• Efficient Reasoning
• Large Language Models