Constrained Decoding

Follow this topic Ask about this topic

Constrained decoding guides a model toward outputs that obey requirements such as a JSON schema, grammar, API specification, code constraints, or permitted label set. Common approaches use finite automata, parsers, tries, and token masking, while search, sampling, verification, and diffusion-based methods handle more complex or global constraints. These techniques can make outputs reliably valid and improve task-specific measures, but validity alone does not ensure correct meaning or successful tool use. A central tension is how to enforce constraints without distorting the model’s probabilities, hindering reasoning, or creating failures when constraints are incomplete.

Latest papers

67 papers16 papers in the last four weeks, up 300% on the four weeks before. 0.1% of all new papers.

When All Rollouts Fail: Policy-Compatible Joint Decoding for Hard-Problem Reasoning
Oct 27, 2026Zehao Wang, Hanzhu Chen, Runquan Gui +2RL for Language Models RL for Language Model Reasoning
University of Science and Technology of China · Tencent · Shanghai Artificial Intelligence Laboratory
Constrained Decoding: Harmful or Beneficial? Revisiting Its Impact on LLM Performance
Oct 26, 2026Ngoc Trinh Hung Nguyen, Alonso Silva, Armen Aghasaryan +1Constrained Decoding Constraint-Guided Reasoning in Language Models
Télécom Paris, Institut Polytechnique de Paris, France · Nokia Bell Labs
DConstrain: Compiling Context-Parameterized Transitions for Diffusion Language Model Constrained Decoding
Oct 22, 2026Shidong Cao, Jing Ma, Wenhao Chai +3Constrained Decoding Grammar-Constrained Decoding
Hong Kong Baptist University · Princeton University · Ant Group
Metamodel-Guided LLM Decoding for Consistent DSL Generation
Oct 21, 2026Boqi Chen, Aren A. Babikian, José Antonio Hernández LópezConstrained Decoding Software Engineering
University of Ottawa, Ottawa, Canada · McGill University, Montreal, Canada · University of Murcia, Murcia, Spain
From Scores to Safe Sets: Exact Safety--Constrained Decoding for Medication Recommendation
Oct 21, 2026Animesh Agarwal, Meysam Ghaffari, Nina Fatehi +2Medication Recommendation Constrained Decoding
Optum
When Copying Is Hard: Copy-Constrained Decoding for Exact Span Reproduction
Oct 20, 2026Jinghui Zhang, Lang Gao, Zongfang Liu +5Language Model Decoding Constrained Decoding
MBZUAI · Zhejiang University · Westlake University +4
When Should Enterprise Policy Decisions Be Learned, and When Should They Be Reasoned?
Oct 13, 2026Hyojung Han, Ryangkyung Kang, Youngeun Kim +1LLM Decision-Making LLM Reasoning
ThakiCloud · Korea University
Structural Automata for Neural Semantic Parsing
Oct 9, 2026Xiao ZhangSemantic Parsing Constrained Decoding
University of Groningen
Breaking the Space Barrier and its Application to Language Model Inference
Oct 6, 2026Arip AsadulaevSchema-Constrained Generation Constrained Decoding
MBZUAI
A Safe Action Is Not Enough: Feasible-Future Decoding for Vision-Language-Action Policies
Oct 4, 2026Tu Nguyen, Matthieu Zimmer, Vu Anh Vu +4Constrained Decoding Vision-Language-Action Models
Huawei Heisenberg Research Center · Huawei Noah’s Ark Lab · TU Berlin +1
Constraint-Aware Training
Oct 2, 2026Jinwoo KimConstrained Generative Modeling Efficient Language Model Training
Department of Computer Science and Engineering University of California-San Diego La Jolla, CA
GrammarRL: Effective Grammar-Constrained Decoding via Reinforcement Learning
Sep 30, 2026Gabriele Tuccio, Antonino Furnari, Aldo Gangemi +1RL for Language Models Constrained Decoding
University of Catania · ISTC - National Research Council, Italy · University of Bologna
Making Grid Beam Search Less Greedy
Sep 30, 2026Sean Papay, Roman KlingerConstrained Decoding Constrained Sampling
Fundamentals of Natural Language Processing University of Bamberg Germany
Controlled Decoding Attacks on Black-Box LLMs
Sep 29, 2026Jesson Wang, Shawn Li, Wei Yang +4Constrained Decoding Adversarial Attacks on LLMs
University of Southern California · Adobe · Amazon
Koa-decision: Fast and Consistent Structured Decision Making with Generative LLMs
Sep 28, 2026Shenghong Dai, Shiva Kumar Pentyala, Yingchi Liu +6LLM Decision-Making LLM Inference Acceleration
Salesforce AI · University of Wisconsin–Madison
Twist, Don't Tilt: Trajectory-Exact Constrained Decoding for Masked Diffusion Models
Sep 28, 2026Aditya Thimmaiah, Lara Marinov, Jayanth Srinivasa +3Constrained Decoding Masked Diffusion Models
The University of Texas at Austin · Cisco Research
Semantic Prefix Oracles for LLM Decoding: Contracts and Differential Validation
Sep 28, 2026Paul Kronlund-DrouaultConstrained Decoding LLM-Based Program Synthesis
ENS de Lyon Lyon, France · Unsuspicious Industries Paris, France · Université de Lille Lille, France
Composable Decoding on the Probability Simplex: Theory and Implementation
Sep 28, 2026Xiaotong Ji, Ahmed Khaled Khamis, Rasul Tutunov +2Constrained Decoding Language Model Decoding
Huawei Noah’s Ark Lab · UCL Centre for AI
Constraints Are Graphs, Not Chains: Exact Decoding for Diffusion Language Models
Sep 26, 2026Jianchang Su, Wei ZhangConstrained Decoding Diffusion Language Model Decoding
University of Connecticut
PTC-Decoder: Towards Intelligent SLMs on Offline Resource-Constrained Edge Devices
Sep 25, 2026Minghui Yu, Ke Mu, Gang WuLLM Tool Use Small Language Models
Shanghai Jiao Tong University, Computer Institute · Shanghai Landfun Information Technology
Constrained Decoding Eliminates Structural Failures in Small LLMs but Reveals a Scale-Dependent Semantic Gap
Sep 20, 2026Akash ChavanConstrained Decoding LLM Evaluation
CLAMP: Constrained Decoding for Vision-Language Embodied Planning
Sep 8, 2026Tianyi Ma, Parisa KordjamshidiConstrained Decoding Vision-Language Grounding
Michigan State University
Confident but Wrong: A Constrained Decoding Diagnostic for Low-Resource Automatic Post-Editing
Aug 30, 2026Isuru Wijesiri, Nisansa de Silva, Kavindu Warnakulasuriya +2Low-Resource MT MT Evaluation
WSO2, Sri Lanka · Dept. of CSE, University of Moratuwa, Sri Lanka · National University of Singapore +1
Stay Within Your Bounds: Distance-Guided Decoding for Guaranteed Context-Free Grammar Compliance
Aug 28, 2026Vincenzo Collura, Karim Tit, Eleonora Giunchiglia +2Constrained Decoding Grammar-Constrained Decoding
University of Luxembourg, Luxembourg · Imperial College London, United Kingdom
Meta-Ctrl: Guaranteed Plan Generation by Decoupling Syntactic and Semantic Constraints
Aug 23, 2026Gwen Yidou-Weng, Edward Sun, Tianyi Ma +5Large Language Model-Based Robot Planning Constrained Decoding
University of California, Los Angeles · Michigan State University
Trie Automata for Constrained Decoding over Large Finite Sets
Aug 12, 2026Xingzi Xu, Karim BouyarmaneConstrained Decoding LLM Inference Acceleration
Amazon
The Parser Already Knows: Lightweight Bias Correction in Constrained Decoding
Aug 10, 2026Işıl Özgü, Yaoxuan Wu, Guy Van den Broeck +1Constrained Decoding Grammar-Constrained Decoding
University of California, Los Angeles Los Angeles, CA, USA
Stateful CARS: Exact Cross-History Reuse for Policy-Constrained LLM Agents
Aug 8, 2026Ibne Farabi Shihab, Md Najmus Swaqeeb, Abu Sa-Adat Mohamed Moon-Im Al AhsanConstrained Decoding LLM Agents
Department of Computer Science, Iowa State University · Department of Computer Science & Engineering, BRAC University
Explicit Note-Event Tokenization and Pitch-Validity Constrained Decoding for MIDI-to-Tablature Transcription
Jul 29, 2026Ting-Kai Hsu, Wei-Chin Wang, Kai-Xi Hong +1Constrained Decoding Automatic Music Transcription
Graduate Institute of Communication Engineering National Taiwan University Taipei, Taiwan · Department of Electrical Engineering National Taiwan University Taipei, Taiwan · Graduate Institute of Networking and Multimedia National Taiwan University Taipei, Taiwan
StageGuard: Physiologically Constrained Sleep Staging
Jul 25, 2026Juntang Wang, Yihan Wang, Hao Wu +3Constrained Decoding Sleep Stage Classification
Zu Chongzhi Center, Duke Kunshan University Kunshan, China

Oct 27, 2026J/K move · Enter open · S save

When All Rollouts Fail: Policy-Compatible Joint Decoding for Hard-Problem Reasoning

Zehao Wang, Hanzhu Chen, Runquan Gui, Zhe Wang+1

University of Science and Technology of China · Tencent · Shanghai Artificial Intelligence Laboratory

RL for Language Models RL for Language Model Reasoning Reinforcement Learning with Verifiable Rewards Constrained Decoding Constraint-Guided Reasoning in Language Models

Open paper AskPDF arXiv

Reinforcement Learning with Verifiable Rewards (RLVR) derives its learning signal from binary rewards via on-policy exploration and becomes a powerful paradigm for improving the reasoning of large language models. Yet it faces a learning cliff on problems beyond model's current ability: when every rollout fails, the samples reveal nothing about a correct solution, and group-relative estimators (e.g., GRPO, RLOO) collapse the update to zero advantage. Expert trajectories offer a remedy, but they are incompatible with the current policy and often lie in the student's low-probability regions, so the surviving updates demand large parameter shifts and decay into surface-level imitation rather than transferable reasoning. We argue that a useful learning signal on hard problems must be both verifiably correct and sufficiently compatible with the current policy. We propose Policy-Compatible Reinforcement Learning (PICO), which performs distribution-constrained joint decoding: expert data guides decoding at the token level while preserving the student's relative token probabilities, and a dynamic top-k mechanism confines the sampling distribution under a finite divergence budget, yielding trajectories that are verified and policy-compatible by construction. These trajectories are optimized jointly with on-policy rollouts under a mixed-behavior objective that mitigates their distribution mismatch via importance sampling. Experiments across reasoning diverse datasets and agentic benchmarks demonstrate that PICO consistently outperforms standard RLVR and prior off-policy baselines.