paper-with-me

Papers

Adaptive Nucleus Truncation for Long-Form Reasoning

2026-06-12 · Ousmane Amadou Dia arxiv

Sampling plays an important role in long-form language-model reasoning. Over thousands of decoding steps, small changes in the candidate token set can compound into different reasoning trajectories, stability profiles, and final answers. Existing truncation methods such as top-$p$, min-$p$, and fixed top-$nσ$ sampling improve over unrestricted sampling, but they rely on fixed thresholds that cannot adapt to changes in entropy, task difficulty, training stage, or generation budget. We introduce Adaptive Nucleus Truncation Sampling (ANTS), which extends top-\(nσ\) sampling from a fixed decoding rule into an adaptive rollout-control mechanism for long-form generation. ANTS selects standardized neighborhoods around the maximum logit before temperature scaling, adapts the truncation width using an entropy-conditioned controller, and retains a no-truncation fallback arm to stabilize training when truncation becomes unsafe. On a 33B-total / 4B-active sparse Mixture-of-Experts reasoning model, ANTS improves average performance over percentage-based benchmarks by +1.9, +3.8, and +5.2 points at 8K, 16K, and 32K generation budgets, respectively. The strongest gains appear on instruction following and mathematical reasoning, with IFBench improving by more than 10 points at 32K and AIME 2025 improving by 7 points. Code generation reveals an important budget interaction. On Codeforces, ANTS trails the baseline at 8K, but reverses this gap and substantially improves ELO at 16K and 32K. These results suggest that sampler design should be treated not just as a decoding hyperparameter, but as part of how we stabilize and scale long-budget reasoning.

📄 PDF Abstract BibTeX arXiv:2606.13982

Code (0)

등록된 구현이 없습니다.

Tasks

Mathematical ReasoningInstruction FollowingCode Generation

Similar Papers 제목 키워드 기반

Reasoning Efficiently Through Adaptive Chain-of-Thought Compression: A Self-Optimizing Framework

2025-09-17 · Kerui Huang, Shuhan Liu, Xing Hu, Tongtong Xu 외 arxiv

Chain-of-Thought (CoT) reasoning enhances Large Language Models (LLMs) by prompting intermediate steps, improving accuracy and robustness in arithmetic, logic, and commonsense tasks. However, this benefit comes with high…

Code Generation

DART: Difficulty-Adaptive Reasoning Truncation for Efficient Large Language Models

2025-11-03 · Ruofan Zhang, Bin Xia, Zhen Cheng, Cairen Jian 외 arxiv

Adaptive reasoning is essential for aligning the computational effort of large language models (LLMs) with the intrinsic difficulty of problems. Current chain-of-thought methods boost reasoning ability but indiscriminate…

Reinforcement Learning

Closing the Curious Case of Neural Text Degeneration

2023-10-02 · Matthew Finlayson, John Hewitt, Alexander Koller, Swabha Swayamdipta 외

Despite their ubiquity in language generation, it remains unknown why truncation sampling heuristics like nucleus sampling are so effective. We provide a theoretical explanation for the effectiveness of the truncation sa…

Text Generation

The Truncation Blind Spot: How Decoding Strategies Systematically Exclude Human-Like Token Choices

2026-03-19 · Esteban Garces Arias, Nurzhan Sapargali, Christian Heumann, Matthias Aßenmacher arxiv

Standard decoding strategies for text generation, including top-$k$, nucleus sampling, and contrastive search, select tokens based on likelihood, restricting outputs to high-probability regions. In contrast, human langua…

Text Generation

DLER: Doing Length pEnalty Right - Incentivizing More Intelligence per Token via Reinforcement Learning

2025-10-16 · Shih-Yang Liu, Xin Dong, Ximing Lu, Shizhe Diao 외 arxiv

Reasoning language models such as OpenAI-o1, DeepSeek-R1, and Qwen achieve strong performance via extended chains of thought but often generate unnecessarily long outputs. Maximizing intelligence per token--accuracy rela…

Reinforcement Learning