Efficient Test-Time Scaling of Multi-Step Reasoning by Probing Internal States of Large Language Models (2026.acl-long)
Copied to clipboard
Jingwei Ni, Ekaterina Fadeeva, Tianyi Wu, Mubashara Akhtar, Jiaheng Zhang, Elliott Ash, Markus Leippold, Timothy Baldwin, See-Kiong Ng, Artem Shelmanov, Mrinmaya Sachan
| Challenge: | Existing verification approaches, such as Process Reward Models, are computationally expensive and limited to specific domains. |
| Approach: | They propose a transformer-based probe that uses internal states of frozen LLMs to estimate credibility of reasoning steps during generation. |
| Outcome: | The proposed probes match or exceed PRMs that are up to 810 larger. |
Similar Papers
Step-level Verifier-guided Hybrid Test-Time Scaling for Large Language Models (2025.emnlp-main)
Copied to clipboard
Kaiyan Chang, Yonghao Shi, Chenglong Wang, Hang Zhou, Chi Hu, Xiaoqian Liu, Yingfeng Luo, Yuan Ge, Tong Xiao, JingBo Zhu
| Challenge: | Recent training-based TTS methods, such as continued reinforcement learning, have surged in popularity, while training-free TTS approaches are gradually fading from prominence. |
| Approach: | They propose a fine-grained sequential scaling method guided by process verification that integrates training-free TTS methods with other classical parallel scaling methods at the step level. |
| Outcome: | Experiments on five instruction-tuned large language models (LLMs) show that training-free TTS methods can extend reasoning performance boundaries. |
Hidden States as Early Signals: Step-level Trace Evaluation and Pruning for Efficient Test-Time Scaling (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to speed up parallel scaling have relied on similarity-based or confidence-based pruning, but these signals do not reliably indicate trace quality. |
| Approach: | They propose a pruning framework that evaluates reasoning steps using hidden states and dynamically prunes unpromising traces during generation. |
| Outcome: | The proposed framework reduces end-to-end inference latency by 45%–70% on average compared to self-consistency while improving reasoning accuracy. |
Guided by Gut: Efficient Test-Time Scaling with Reinforced Intrinsic Confidence (2026.acl-long)
Copied to clipboard
| Challenge: | Guided by Gut (GG) is an efficient self-guided TTS framework for Large Language Models (LLMs) that performs step-by-step reasoning at a low cost without any reward models or verifiers. |
| Approach: | They propose a self-guided TTS framework that enables LLMs to perform step-by-step reasoning at a low cost without any reward models or verifiers. |
| Outcome: | Empirical evaluations show that GG performs better than TTS with PRMs while reducing GPU memory usage by up to 10. |
Beyond Memorization: Extending Reasoning Depth with Recurrence, Memory and Test-Time Compute Scaling (2026.findings-acl)
Copied to clipboard
Ivan Rodkin, Daniil Orel, Konstantin Smirnov, Arman Bolatov, Bilal Elbouardi, Besher Hassan, Yuri Kuratov, Aydar Bulatov, Preslav Nakov, Timothy Baldwin, Artem Shelmanov, Mikhail Burtsev
| Challenge: | Reasoning is a core capability of large language models, yet how multi-step reasoning is learned and executed remains unclear. |
| Approach: | They evaluate how large language models learn multi-step reasoning without memorization . they find that most neural architectures trained from scratch can learn rule inference . |
| Outcome: | The proposed framework fails to solve a natural-language proxy task with high accuracy. |
Towards Hierarchical Multi-Step Reward Models for Enhanced Reasoning in Large Language Models (2026.findings-acl)
Copied to clipboard
Teng Wang, Jiang Zhangyi, Zhenqi He, Hailei Gong, Shenyang Tong, Wenhan Yang, Zeyu Li, Yanan Zheng, Zifan He, Zewen Ye, Shengjie Ma, Jianping Zhang
| Challenge: | Existing Process Reward Models (PRMs) are vulnerable to reward hacking and require expensive, large-scale annotation of reasoning steps. |
| Approach: | They propose a reward model approach which evaluates both individual and consecutive reasoning steps from fine-grained and coarse-grounded level. |
| Outcome: | Empirical results show that the proposed model performs better than existing PRMs and is more robust than existing models. |
a1: Steep Test-time Scaling Law via Environment Augmented Generation (2026.findings-acl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have made remarkable advances in reasoning, yet continue to struggle with hallucinations, logical errors, and inability to self-correct during complex multi-step tasks. |
| Approach: | They propose a framework that enhances LLM reasoning through real-time environmental feedback validating each reasoning step, dynamic branch exploration for investigating alternative solution paths when faced with errors, and experience-based learning from successful reasoning trajectories. |
| Outcome: | The proposed model outperforms comparable models by 24.4 percentage points across benchmarks while outperforming comparable models. |
Stepwise Reasoning Checkpoint Analysis: A Test Time Scaling Method to Enhance LLMs’ Reasoning (2025.emnlp-main)
Copied to clipboard
Zezhong Wang, Xingshan Zeng, Weiwen Liu, Yufei Wang, Liangyou Li, Yasheng Wang, Lifeng Shang, Xin Jiang, Qun Liu, Kam-Fai Wong
| Challenge: | Existing methods that use Chain-of-Thought suffer from path homogenization and inefficient use of intermediate results. |
| Approach: | They propose a framework that introduces checkpoints between reasoning steps to reduce path homogenization and create fault-tolerant mechanisms. |
| Outcome: | The proposed framework reduces path homogenization and creates fault-tolerant mechanism by utilizing high-quality intermediate results. |
Prompting Test-Time Scaling Is A Strong LLM Reasoning Data Augmentation (2026.findings-acl)
Copied to clipboard
| Challenge: | Large language models exhibit strong reasoning when guided by chain-of-thought exemplars . collecting large, high-quality reasoning datasets remains laborious and resource-intensive . |
| Approach: | They propose a prompt-space data augmentation framework for enhancing LLM reasoning . they use a pool of 90 randomly selected reasoning instances to elicit diverse reasoning trajectories . |
| Outcome: | The proposed framework improves accuracy over small-data benchmarks and generalization on out-of-domain reasoning evaluations. |
Beyond Outcome Verification: Verifiable Process Reward Models for Structured Reasoning (2026.findings-acl)
Copied to clipboard
| Challenge: | Recent work on reinforcement learning with verifiable rewards (RLVR) has shown that large language models can be substantially improved using outcome-level verification signals. |
| Approach: | They propose a framework where intermediate reasoning steps are checked by deterministic, rule-based verifiers. |
| Outcome: | The proposed framework achieves 20% higher F1 than state-of-the-art models and 6.5% higher than verifiable outcome rewards, with substantial gains in evidence grounding and logical coherence. |
FineReason: Evaluating and Improving LLMs’ Deliberate Reasoning through Reflective Puzzle Solving (2025.acl-long)
Copied to clipboard
Guizhen Chen, Weiwen Xu, Hao Zhang, Hou Pong Chan, Chaoqun Liu, Lidong Bing, Deli Zhao, Anh Tuan Luu, Yu Rong
| Challenge: | Recent advances in large language models (LLMs) highlight an important shift from the “System 1” way of quick reactions to the “system 2” style of reflection-and-correction problem solving. |
| Approach: | They propose a logic-puzzle benchmark for systematic evaluation of large language models' reasoning capabilities that decomposes each puzzle into atomic steps. |
| Outcome: | The proposed model improves on state checking and state transition tasks and demonstrates gains in reasoning by up to 5.1%. |