Tag: Preference Training

All the papers with the tag "Preference Training".

• AdaR1: From Long-CoT to Hybrid-CoT via Bi-Level Adaptive Reasoning Optimization

  grok-3-latest

  Score: 0.74

  Published:2025年4月30日 at  at 14:01

  #LLM, #Reasoning, #Efficiency Optimization, #Model Merging, #Preference Training

  本文提出 AdaR1 框架，通过模型融合和双层偏好训练实现自适应推理，显著降低大型语言模型推理成本（平均长度减少超 50%）同时保持高性能。