Saved in:
Bibliographic Details
Main Authors: Li, Xianzhi, Callanan, Ethan, Ghassel, Abdellah, Zhu, Xiaodan
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2503.21961
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915756997869568
author Li, Xianzhi
Callanan, Ethan
Ghassel, Abdellah
Zhu, Xiaodan
author_facet Li, Xianzhi
Callanan, Ethan
Ghassel, Abdellah
Zhu, Xiaodan
contents Test-time compute methods can significantly improve the reasoning capabilities and problem-solving accuracy of large language models (LLMs). However, these approaches require substantially more computational resources, with most compute wasted on exploring low-diversity branches where the model already exhibits high confidence. We observe that a small subset of uncertain reasoning steps has a disproportionately large impact on final prediction accuracy, and branching at these critical junctures tends to yield more diverse and higher-quality candidate reasoning steps. We propose Entropy-Gated Branching (EGB), which branches only at high-uncertainty steps and prunes expansions with a lightweight verifier. On mathematical and financial reasoning benchmarks, EGB improves accuracy by 22.6% over standard inference while operating 31%-75% faster across math benchmarks than test-time beam search with higher performance. Our results show that dynamic resource allocation during inference can substantially improve both efficiency and effectiveness, offering a more scalable pathway to enhanced LLM reasoning capabilities.
format Preprint
id arxiv_https___arxiv_org_abs_2503_21961
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Entropy-Gated Branching for Efficient Test-Time Reasoning
Li, Xianzhi
Callanan, Ethan
Ghassel, Abdellah
Zhu, Xiaodan
Computation and Language
Artificial Intelligence
Test-time compute methods can significantly improve the reasoning capabilities and problem-solving accuracy of large language models (LLMs). However, these approaches require substantially more computational resources, with most compute wasted on exploring low-diversity branches where the model already exhibits high confidence. We observe that a small subset of uncertain reasoning steps has a disproportionately large impact on final prediction accuracy, and branching at these critical junctures tends to yield more diverse and higher-quality candidate reasoning steps. We propose Entropy-Gated Branching (EGB), which branches only at high-uncertainty steps and prunes expansions with a lightweight verifier. On mathematical and financial reasoning benchmarks, EGB improves accuracy by 22.6% over standard inference while operating 31%-75% faster across math benchmarks than test-time beam search with higher performance. Our results show that dynamic resource allocation during inference can substantially improve both efficiency and effectiveness, offering a more scalable pathway to enhanced LLM reasoning capabilities.
title Entropy-Gated Branching for Efficient Test-Time Reasoning
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2503.21961