ACL Findings2026
Beyond Templates: Dynamic Adaptation of Reasoning Demonstrations via Feasibility-Aware Exploration
Yong Wu, Weihang Pan, Ke Li, Chen Binhui, Ping Li, Binbin Lin
Abstract
Large language models (LLMs) have shown remarkable reasoning capabilities, yet aligning such abilities to small language models (SLMs) remains a challenge due to distributional mismatches and limited model capacity. Existing reasoning datasets, typically designed for powerful LLMs, often lead to degraded performance when directly applied to weaker models. In this work, we introduce Dynamic Adaptation of Reasoning Trajectories (DART), a novel data adaptation framework that bridges the capability gap between expert reasoning trajectories and diverse SLMs. Instead of uniformly imitating expert steps, DART employs a selective imitation strategy guided by step-wise adaptability estimation via solution simulation. When expert steps surpass the student's capacity-signaled by an imitation gap-the student autonomously explores alternative reasoning paths, constrained by outcome consistency. We validate DART across multiple reasoning benchmarks and model scales, demonstrating that it significantly improves generalization and data efficiency over static fine-tuning. Our method enhances supervision quality by aligning training signals with the student's reasoning capabilities, offering a scalable solution for reasoning alignment in resource-constrained models. Our code is available at https://github.com/yw- 0311/BeyondTemplates.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
Lineage
- Rubrics as Rewards: Reinforcement Learning Beyond Verifiable Domains cites Let's Verify Step by Step
- Let's Verify Step by Step cites Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
- MAmmoTH2: Scaling Instructions from the Web cites Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
- MAmmoTH2: Scaling Instructions from the Web cites MetaMath: Bootstrap Your Own Mathematical Questions for Large Language Models
- MetaMath: Bootstrap Your Own Mathematical Questions for Large Language Models cites Let's Verify Step by Step
- MetaMath: Bootstrap Your Own Mathematical Questions for Large Language Models cites Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
Builds on10
- Chain-of-Thought Prompting Elicits Reasoning in Large Language ModelsJason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma et al.NeurIPS 2022 · 22,562 citations
- Let's Verify Step by StepHunter Lightman, Vineet Kosaraju, Yuri Burda, Harrison Edwards et al.ICLR 2024 · 3,045 citations
- MetaMath: Bootstrap Your Own Mathematical Questions for Large Language ModelsLonghui Yu, Weisen Jiang, Han Shi, Jincheng Yu et al.ICLR 2024 · 637 citations
- Rubrics as Rewards: Reinforcement Learning Beyond Verifiable DomainsAnisha Gunjal, Anthony Wang, Elaine Lau, Vaskar Nath et al.ICLR 2026 · 340 citations
- MAmmoTH2: Scaling Instructions from the WebXiang Yue, Tianyu Zheng, Ge Zhang, Wenhu ChenNeurIPS 2024 · 176 citations
Related papers
- SmartAD: Capacity-Aligned Agent Distillation for Small Language ModelsGuokai Tang, Feng ZhaoACL Findings 2026
- Making Expert Reasoning Learnable with Self-DistillationEthan Mendes, Jungsoo Park, Alan RitterICML 2026 · 1 citation
- LightReasoner: Can Small Language Models Teach Large Language Models Reasoning?Jingyuan Wang, Yankai Chen, Zhonghang Li, Chao HuangACL 2026
- TARo: Token-level Adaptive Routing for LLM Test-time AlignmentArushi Rai, Qiang Zhang, Hanqing Zeng, Yunkai Zhang et al.ACL Findings 2026
- DOTS: Learning to Reason Dynamically in LLMs via Optimal Reasoning Trajectories SearchMurong Yue, Wenlin Yao, Haitao Mi, Dian Yu et al.ICLR 2025
