强化学习不再靠奖励？组合优化迎来“偏好驱动”新框架
"偏好优化" 相关文章
Snowflake Proposes ExCoT: A Novel AI Framework that Iteratively Optimizes Open-Source LLMs by Combining CoT Reasoning with off-Policy and on-Policy DPO, Relying Solely on Execution Accuracy as Feedback
给语音模型戴上「眼镜」，错误率降低12.5%，人大CMU最新开源
DPO-Shift：一个参数可控改变DPO分布，缓解似然偏移
This AI Paper from Meta Introduces Diverse Preference Optimization (DivPO): A Novel Optimization Method for Enhancing Diversity in Large Language Models
Meta AI Proposes EvalPlanner: A Preference Optimization Algorithm for Thinking-LLM-as-a-Judge
Test-Time Preference Optimization: A Novel AI Framework that Optimizes LLM Outputs During Inference with an Iterative Textual Reward Policy
o1也会「想太多」？腾讯AI Lab与上海交大揭秘o1模型过度思考问题
社区供稿 | 引入隐式模型融合技术，中山大学团队推出 FuseChat-3.0
【NLP】万字长文梳理LLM+RL(HF)的脉络
LongAlign: A Segment-Level Encoding Method to Enhance Long-Text to Image Generation
HyPO: A Hybrid Reinforcement Learning Algorithm that Uses Offline Data for Contrastive-based Preference Optimization and Online Unlabeled Data for KL Regularization
为视觉语言多模态模型进行偏好优化
This AI Paper from Cohere for AI Presents a Comprehensive Study on Multilingual Preference Optimization
MaPO: The Memory-Friendly Maestro – A New Standard for Aligning Generative Models with Diverse Preferences
