mlsanityData to 7 Oct 2026

Adaptive Decoding via Test-Time Policy Learning for Self-Improving Generation

arXiv 2603.18428 ↗First posted 19 Mar 2026cs.CL cs.AIcited by 0

On the map: Synthetic data and self-improvement for LLMs → Self-rewarding alignment and self-training

Abstract (arXiv)

Decoding strategies largely determine the quality of Large Language Model (LLM) outputs, yet widely used heuristics such as greedy or fixed temperature/top-p decoding are static and often task-agnostic, leading to suboptimal or inconsistent generation quality across domains that demand stylistic or structural flexibility. We introduce a reinforcement learning-based decoder sampler that treats decoding as sequential decision-making and learns a lightweight policy to adjust sampling parameters at test-time while keeping LLM weights frozen. We evaluated summarization datasets including BookSum, arXiv, and WikiHow using Granite-3.3-2B and Qwen-2.5-0.5B. Our policy sampler consistently outperforms greedy and static baselines, achieving relative gains of up to +88% (BookSum, Granite) and +79% (WikiHow, Qwen). Reward ablations show that overlap-only objectives underperform compared to composite rewards, while structured shaping terms (length, coverage, repetition, completeness) enable stable and sustained improvements. These findings highlight reinforcement learning as a practical mechanism for test-time adaptation in decoding, enabling domain-aware and user-controllable generation without retraining large models.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Self-rewarding alignment and self-training, one of 13 directions in Synthetic data and self-improvement for LLMs. Builds on 10 of its 15 references that are in the map. First posted on arXiv 19 Mar 2026.

Builds on

10 of 15 references.

References · 10

  1. 2020Language Models are Few-Shot Learnerscited by 32,199
  2. 2019The Curious Case of Neural Text Degenerationcited by 2,881
  3. 2019SuperGLUE: A Stickier Benchmark for General-Purpose Language Understanding Systemscited by 1,718
  4. 2018Hierarchical Neural Story Generationcited by 1,182
  5. 2023A Comprehensive Overview of Large Language Modelscited by 639
  6. 2022Contrastive Decoding: Open-ended Text Generation as Optimizationcited by 598
  7. 2023Harnessing the Power of LLMs in Practice: A Survey on ChatGPT and Beyondcited by 409
  8. 2022A Survey of Controllable Text Generation using Transformer-based Pre-trained Language Modelscited by 128
  9. 2024A Thorough Examination of Decoding Methods in the Era of LLMscited by 112
  10. 2025Speculative Decoding and Beyond: An In-Depth Survey of Techniquescited by 22

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Learning to Draft: Adaptive Speculative Decoding with Reinforcement Learning
  2. 2026Reinforcement Inference: Leveraging Uncertainty for Self-Correcting Language Model Reasoning
  3. 2025On the Limits of Test-Time Compute: Sequential Reward Filtering for Better Inference
  4. 2025Optimal Stopping vs Best-of-N for Inference Time Optimization
  5. 2026DecompRL: Solving Harder Problems by Learning Modular Code Generation
  6. 2026Entropy-informed Decoding: Adaptive Information-Driven Branching
  7. 2026CodeScaler: Scaling Code LLM Training and Test-Time Inference via Reward Models
  8. 2025Let it Calm: Exploratory Annealed Decoding for Verifiable Reinforcement Learning