ParaTempo: Efficient Parallel Reasoning via Temporal Confidence

2026-08-17

AI Summary

This study addresses the high computational costs and challenges of dynamic branch management in parallel inference by proposing ParaTempo, a training-free asynchronous framework. ParaTempo innovatively introduces a temporal confidence metric to assess branch convergence, enabling synchronization-free adaptive pruning, early stopping, and resource reallocation. Evaluated on mathematical and scientific reasoning tasks, the method maintains competitive accuracy while reducing average latency by 21.8%–32.2% and token consumption by 18.1%–30.3%. These results demonstrate that ParaTempo significantly enhances both the efficiency and resource utilization of asynchronous parallel inference for large language models, offering a practical solution to the scalability bottlenecks inherent in current parallel decoding paradigms without requiring additional model training or architectural modifications.

Abstract

Parallel reasoning improves the accuracy and robustness of large reasoning models by exploring multiple solution paths, but its computational cost grows with reasoning depth and branch count. Existing methods for managing these parallel paths typically rely on final-answer consensus, local token confidence, or isolated intermediate probes. However, these signals are often delayed, weakly tied to actual reasoning progress, or too noisy for dynamic, branch-level control. To address these limitations, we introduce ParaTempo, a training-free asynchronous parallel reasoning framework. ParaTempo is driven by temporal confidence, a branch-local measure of answer-space convergence. Each branch is periodically probed for a tentative answer probability distribution, and temporal confidence quantifies how sharply the recent intermediate probes concentrate on a dominant answer. Once sufficient evidence has accumulated, ParaTempo drives its entire control process from this single signal: low-confidence branches are pruned, branches that persistently commit to their dominant answer are retired early, freed computation is reallocated by forking new branches, and generation stops globally once the confidence-weighted vote concentrates. Without requiring synchronization among reasoning trajectories, ParaTempo adaptively allocates computation based on branch-level convergence. Experiments on challenging mathematical and scientific reasoning benchmarks show that ParaTempo reduces average latency by 21.8-32.2% and total token usage by 18.1-30.3% while maintaining competitive accuracy. Moreover, temporal confidence exhibits stronger temporal stability and predictive power for future branch convergence than token-level and instantaneous signals.

Problem

• Parallel Reasoning
• Computational Efficiency
• Dynamic Branch Control
• Large Reasoning Models

Innovation

• Temporal Confidence
• Asynchronous Parallel Reasoning
• Training-free Framework
• Adaptive Computation Allocation
• Branch-level Convergence