Saved in:
Bibliographic Details
Main Authors: Jiang, Wenxuan, Zuo, Yuxin, Zhang, Zijian, Wu, Xuecheng, Fan, Zining, Liu, Wenxuan, Chen, Li, Li, Xiaoyu, Cao, Xuezhi, Jin, Xiaolong, Liu, Ninghao
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2604.00438
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917376825491456
author Jiang, Wenxuan
Zuo, Yuxin
Zhang, Zijian
Wu, Xuecheng
Fan, Zining
Liu, Wenxuan
Chen, Li
Li, Xiaoyu
Cao, Xuezhi
Jin, Xiaolong
Liu, Ninghao
author_facet Jiang, Wenxuan
Zuo, Yuxin
Zhang, Zijian
Wu, Xuecheng
Fan, Zining
Liu, Wenxuan
Chen, Li
Li, Xiaoyu
Cao, Xuezhi
Jin, Xiaolong
Liu, Ninghao
contents In-Context Reinforcement Learning (ICRL) enables Large Language Models (LLMs) to learn online from external rewards directly within the context window. However, a central challenge in ICRL is reward estimation, as models typically lack access to ground-truths during inference. To address this limitation, we propose Test-Time Rethinking for In-Context Reinforcement Learning (TR-ICRL), a novel ICRL framework designed for both reasoning and knowledge-intensive tasks. TR-ICRL operates by first retrieving the most relevant instances from an unlabeled evaluation set for a given query. During each ICRL iteration, LLM generates a set of candidate answers for every retrieved instance. Next, a pseudo-label is derived from this set through majority voting. This label then serves as a proxy to give reward messages and generate formative feedbacks, guiding LLM through iterative refinement. In the end, this synthesized contextual information is integrated with the original query to form a comprehensive prompt, with the answer determining through a final round of majority voting. TR-ICRL is evaluated on mainstream reasoning and knowledge-intensive tasks, where it demonstrates significant performance gains. Remarkably, TR-ICRL improves Qwen2.5-7B by 21.23% on average on MedQA and even 137.59% on AIME2024. Extensive ablation studies and analyses further validate the effectiveness and robustness of our approach. Our code is available at https://github.com/pangpang-xuan/TR_ICRL.
format Preprint
id arxiv_https___arxiv_org_abs_2604_00438
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle TR-ICRL: Test-Time Rethinking for In-Context Reinforcement Learning
Jiang, Wenxuan
Zuo, Yuxin
Zhang, Zijian
Wu, Xuecheng
Fan, Zining
Liu, Wenxuan
Chen, Li
Li, Xiaoyu
Cao, Xuezhi
Jin, Xiaolong
Liu, Ninghao
Computation and Language
In-Context Reinforcement Learning (ICRL) enables Large Language Models (LLMs) to learn online from external rewards directly within the context window. However, a central challenge in ICRL is reward estimation, as models typically lack access to ground-truths during inference. To address this limitation, we propose Test-Time Rethinking for In-Context Reinforcement Learning (TR-ICRL), a novel ICRL framework designed for both reasoning and knowledge-intensive tasks. TR-ICRL operates by first retrieving the most relevant instances from an unlabeled evaluation set for a given query. During each ICRL iteration, LLM generates a set of candidate answers for every retrieved instance. Next, a pseudo-label is derived from this set through majority voting. This label then serves as a proxy to give reward messages and generate formative feedbacks, guiding LLM through iterative refinement. In the end, this synthesized contextual information is integrated with the original query to form a comprehensive prompt, with the answer determining through a final round of majority voting. TR-ICRL is evaluated on mainstream reasoning and knowledge-intensive tasks, where it demonstrates significant performance gains. Remarkably, TR-ICRL improves Qwen2.5-7B by 21.23% on average on MedQA and even 137.59% on AIME2024. Extensive ablation studies and analyses further validate the effectiveness and robustness of our approach. Our code is available at https://github.com/pangpang-xuan/TR_ICRL.
title TR-ICRL: Test-Time Rethinking for In-Context Reinforcement Learning
topic Computation and Language
url https://arxiv.org/abs/2604.00438