Adaptive Termination for Multi-round Parallel Reasoning: An Universal Semantic Entropy-Guided Framework
This addresses a key bottleneck in scaling inference for AI systems, offering a practical improvement for researchers and developers, though it is incremental as it builds on existing reasoning paradigms.
The paper tackles the inefficiency and lack of coordination in sequential and parallel reasoning for large language models by introducing semantic entropy, a metric that guides dynamic termination in a collaborative inference framework, achieving up to 40% reduction in computational cost while maintaining accuracy.
Recent advances in large language models (LLMs) have accelerated progress toward artificial general intelligence, with inference-time scaling emerging as a key technique. Contemporary approaches leverage either sequential reasoning (iteratively extending chains of thought) or parallel reasoning (generating multiple solutions simultaneously) to scale inference. However, both paradigms face fundamental limitations: sequential scaling typically relies on arbitrary token budgets for termination, leading to inefficiency or premature cutoff; while parallel scaling often lacks coordination among parallel branches and requires intrusive fine-tuning to perform effectively. In light of these challenges, we aim to design a flexible test-time collaborative inference framework that exploits the complementary strengths of both sequential and parallel reasoning paradigms. Towards this goal, the core challenge lies in developing an efficient and accurate intrinsic quality metric to assess model responses during collaborative inference, enabling dynamic control and early termination of the reasoning trace. To address this challenge, we introduce semantic entropy (SE), which quantifies the semantic diversity of parallel model responses and serves as a robust indicator of reasoning quality due to its strong negative correlation with accuracy...