Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Uscidda, Theo, Trager, Matthew, Kleinman, Michael, Chattopadhyay, Aditya, Xia, Wei, Soatto, Stefano
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2509.20368
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866915511702388736
author Uscidda, Theo
Trager, Matthew
Kleinman, Michael
Chattopadhyay, Aditya
Xia, Wei
Soatto, Stefano
author_facet Uscidda, Theo
Trager, Matthew
Kleinman, Michael
Chattopadhyay, Aditya
Xia, Wei
Soatto, Stefano
contents One common strategy for improving the performance of Large Language Models (LLMs) on downstream tasks involves using a \emph{verifier model} to either select the best answer from a pool of candidates or to steer the auto-regressive generation process towards better outputs. This class of methods typically results in improved accuracy at the cost of increased computation at test-time, a paradigm known as \emph{test-time scaling}. However, most existing approaches increase computation uniformly across all samples and generation steps, without considering the complexity of individual instances, leading to inefficient resource use. We address this limitation by proposing an approach, called \emph{Locally Adaptive Test-Time Scaling (LATTS)}, that allocates variable compute across generation steps. Specifically, at each generation step, LATTS employs a verifier-based acceptance criterion to decide whether to resample, backtrack, restart, or stop the generation process. This criterion effectively adjusts the per-step computational effort based on a precise notion of \emph{local difficulty} derived from the verifier model. Empirical results show that LATTS achieves significantly superior accuracy--compute tradeoffs compared to standard verifier-based methods.
format Preprint
id arxiv_https___arxiv_org_abs_2509_20368
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle LATTS: Locally Adaptive Test-Time Scaling
Uscidda, Theo
Trager, Matthew
Kleinman, Michael
Chattopadhyay, Aditya
Xia, Wei
Soatto, Stefano
Artificial Intelligence
One common strategy for improving the performance of Large Language Models (LLMs) on downstream tasks involves using a \emph{verifier model} to either select the best answer from a pool of candidates or to steer the auto-regressive generation process towards better outputs. This class of methods typically results in improved accuracy at the cost of increased computation at test-time, a paradigm known as \emph{test-time scaling}. However, most existing approaches increase computation uniformly across all samples and generation steps, without considering the complexity of individual instances, leading to inefficient resource use. We address this limitation by proposing an approach, called \emph{Locally Adaptive Test-Time Scaling (LATTS)}, that allocates variable compute across generation steps. Specifically, at each generation step, LATTS employs a verifier-based acceptance criterion to decide whether to resample, backtrack, restart, or stop the generation process. This criterion effectively adjusts the per-step computational effort based on a precise notion of \emph{local difficulty} derived from the verifier model. Empirical results show that LATTS achieves significantly superior accuracy--compute tradeoffs compared to standard verifier-based methods.
title LATTS: Locally Adaptive Test-Time Scaling
topic Artificial Intelligence
url https://arxiv.org/abs/2509.20368