CG-TTRL: Context-Guided Test-Time Reinforcement Learning for On-Device Large Language Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Hosseini, Peyman, Bohdal, Ondrej, Ceritli, Taha, Castro, Ignacio, Purver, Matthew, Ozay, Mete, Michieli, Umberto
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918192380641280
author Hosseini, Peyman
Bohdal, Ondrej
Ceritli, Taha
Castro, Ignacio
Purver, Matthew
Ozay, Mete
Michieli, Umberto
author_facet Hosseini, Peyman
Bohdal, Ondrej
Ceritli, Taha
Castro, Ignacio
Purver, Matthew
Ozay, Mete
Michieli, Umberto
contents Test-time Reinforcement Learning (TTRL) has shown promise in adapting foundation models for complex tasks at test-time, resulting in large performance improvements. TTRL leverages an elegant two-phase sampling strategy: first, multi-sampling derives a pseudo-label via majority voting, while subsequent downsampling and reward-based fine-tuning encourages the model to explore and learn diverse valid solutions, with the pseudo-label modulating the reward signal. Meanwhile, in-context learning has been widely explored at inference time and demonstrated the ability to enhance model performance without weight updates. However, TTRL's two-phase sampling strategy under-utilizes contextual guidance, which can potentially improve pseudo-label accuracy in the initial exploitation phase while regulating exploration in the second. To address this, we propose context-guided TTRL (CG-TTRL), integrating context dynamically into both sampling phases and propose a method for efficient context selection for on-device applications. Our evaluations on mathematical and scientific QA benchmarks show CG-TTRL outperforms TTRL (e.g. additional 7% relative accuracy improvement over TTRL), while boosting efficiency by obtaining strong performance after only a few steps of test-time training (e.g. 8% relative improvement rather than 1% over TTRL after 3 steps).
format Preprint
id arxiv_https___arxiv_org_abs_2511_06430
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle CG-TTRL: Context-Guided Test-Time Reinforcement Learning for On-Device Large Language Models
Hosseini, Peyman
Bohdal, Ondrej
Ceritli, Taha
Castro, Ignacio
Purver, Matthew
Ozay, Mete
Michieli, Umberto
Machine Learning
Computation and Language
I.2.7; I.5.4
Test-time Reinforcement Learning (TTRL) has shown promise in adapting foundation models for complex tasks at test-time, resulting in large performance improvements. TTRL leverages an elegant two-phase sampling strategy: first, multi-sampling derives a pseudo-label via majority voting, while subsequent downsampling and reward-based fine-tuning encourages the model to explore and learn diverse valid solutions, with the pseudo-label modulating the reward signal. Meanwhile, in-context learning has been widely explored at inference time and demonstrated the ability to enhance model performance without weight updates. However, TTRL's two-phase sampling strategy under-utilizes contextual guidance, which can potentially improve pseudo-label accuracy in the initial exploitation phase while regulating exploration in the second. To address this, we propose context-guided TTRL (CG-TTRL), integrating context dynamically into both sampling phases and propose a method for efficient context selection for on-device applications. Our evaluations on mathematical and scientific QA benchmarks show CG-TTRL outperforms TTRL (e.g. additional 7% relative accuracy improvement over TTRL), while boosting efficiency by obtaining strong performance after only a few steps of test-time training (e.g. 8% relative improvement rather than 1% over TTRL after 3 steps).
title CG-TTRL: Context-Guided Test-Time Reinforcement Learning for On-Device Large Language Models
topic Machine Learning
Computation and Language
I.2.7; I.5.4
url https://arxiv.org/abs/2511.06430