Tag: Reasoning

All the articles with the tag "Reasoning".

• Exploring the Potential of Offline RL for Reasoning in LLMs: A Preliminary Study

  Published: 7 May, 2025 |  at  08:41 AM

  

  86.49 🤔

  本文通过探索离线强化学习方法（LD-DPO），在DeepDistill-32B模型上实现了平均3.3%的推理性能提升，尤其在Arena-Hard基准上提升10.1%，并强调了推理长度与语义丰富性平衡的重要性。

• Skip-Thinking: Chunk-wise Chain-of-Thought Distillation Enable Smaller Language Models to Reason Better and Faster

  Published: 1 Jun, 2025 |  at  11:53 AM

  

  86.49 🤔

  本文提出分块训练（CWT）和跳跃思维训练（STT），通过将推理过程分块并跳过非核心块，显著提升小型语言模型在链式思维蒸馏中的推理准确性和速度。

• Can LLMs Maintain Fundamental Abilities under KV Cache Compression?

  Published: 26 May, 2025 |  at  11:22 AM

  

  86.44 🤔

  本文通过KVFundaBench基准系统评估KV缓存压缩对大型语言模型基本能力的影响，揭示任务依赖性性能降解，并提出ShotKV方法，通过区分预填充和解码阶段压缩策略，在长上下文生成任务上显著提升性能。

• Advancing Multimodal Reasoning via Reinforcement Learning with Cold Start

  Published: 3 Jun, 2025 |  at  11:30 AM

  

  86.44 🤔

  本文通过质疑‘aha moment’模式与推理能力提升的相关性，提出了一种结合监督微调（SFT）和强化学习（RL）的两阶段方法，在3B和7B规模的多模态大语言模型上显著提升了多模态推理性能，达到开源模型中的最优水平。

• Do We Truly Need So Many Samples? Multi-LLM Repeated Sampling Efficiently Scales Test-Time Compute

  Published: 10 May, 2025 |  at  10:59 AM

  

  86.42 🤔

  This paper introduces ModelSwitch, a multi-LLM repeated sampling strategy that leverages answer consistency to dynamically switch models, achieving superior performance and 34% sample efficiency over single-LLM self-consistency across diverse datasets.