DynScaling: Efficient Verifier-free Inference Scaling via Dynamic and Integrated Sampling

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wang, Fei, Wan, Xingchen, Sun, Ruoxi, Chen, Jiefeng, Arık, Sercan Ö.
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909653013626880
author Wang, Fei
Wan, Xingchen
Sun, Ruoxi
Chen, Jiefeng
Arık, Sercan Ö.
author_facet Wang, Fei
Wan, Xingchen
Sun, Ruoxi
Chen, Jiefeng
Arık, Sercan Ö.
contents Inference-time scaling has proven effective in boosting large language model (LLM) performance through increased test-time computation. Yet, its practical application is often hindered by reliance on external verifiers or a lack of optimization for realistic computational constraints. We propose DynScaling, which addresses these limitations through two primary innovations: an integrated parallel-sequential sampling strategy and a bandit-based dynamic budget allocation framework. The integrated sampling strategy unifies parallel and sequential sampling by constructing synthetic sequential reasoning chains from initially independent parallel responses, promoting diverse and coherent reasoning trajectories. The dynamic budget allocation framework formulates the allocation of computational resources as a multi-armed bandit problem, adaptively distributing the inference budget across queries based on the uncertainty of previously sampled responses, thereby maximizing computational efficiency. By combining these components, DynScaling effectively improves LLM performance under practical resource constraints without the need for external verifiers. Experimental results demonstrate that DynScaling consistently surpasses existing verifier-free inference scaling baselines in both task performance and computational cost.
format Preprint
id arxiv_https___arxiv_org_abs_2506_16043
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle DynScaling: Efficient Verifier-free Inference Scaling via Dynamic and Integrated Sampling
Wang, Fei
Wan, Xingchen
Sun, Ruoxi
Chen, Jiefeng
Arık, Sercan Ö.
Computation and Language
Artificial Intelligence
Machine Learning
Inference-time scaling has proven effective in boosting large language model (LLM) performance through increased test-time computation. Yet, its practical application is often hindered by reliance on external verifiers or a lack of optimization for realistic computational constraints. We propose DynScaling, which addresses these limitations through two primary innovations: an integrated parallel-sequential sampling strategy and a bandit-based dynamic budget allocation framework. The integrated sampling strategy unifies parallel and sequential sampling by constructing synthetic sequential reasoning chains from initially independent parallel responses, promoting diverse and coherent reasoning trajectories. The dynamic budget allocation framework formulates the allocation of computational resources as a multi-armed bandit problem, adaptively distributing the inference budget across queries based on the uncertainty of previously sampled responses, thereby maximizing computational efficiency. By combining these components, DynScaling effectively improves LLM performance under practical resource constraints without the need for external verifiers. Experimental results demonstrate that DynScaling consistently surpasses existing verifier-free inference scaling baselines in both task performance and computational cost.
title DynScaling: Efficient Verifier-free Inference Scaling via Dynamic and Integrated Sampling
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2506.16043