Semantic-guided Diverse Decoding for Large Language Model

Weijie Shi, Yue Cui, Yaguang Wu, Jingzhi Fang, Shibo Zhang, Mengze Li, Sirui Han, Jia Zhu, Jiajie Xu, Xiaofang Zhou

Keywords: Diverse Decoding, Sampling Strategy, Large Language Model

TL;DR: SemDiD generates semantically diverse LLM outputs by guiding decoding in embedding space through orthogonal direction vectors and inter-group repulsion, outperforming existing methods in Best-of-N evaluations and accelerating RLHF training.

Abstract: Diverse decoding of large language models is crucial for applications requiring multiple semantically distinct responses, yet existing methods primarily achieve lexical rather than semantic diversity. This limitation significantly constrains Best-of-N strategies, group-based reinforcement learning, and data synthesis. While temperature sampling and diverse beam search modify token distributions or apply n-gram penalties, they fail to ensure meaningful semantic differentiation. We introduce Semantic-guided Diverse Decoding (SemDiD), operating directly in embedding space that balances quality with diversity through three complementary mechanisms: orthogonal directional guidance, dynamic inter-group repulsion, and position-debiased probability assessment. SemDiD harmonizes these competing objectives using adaptive gain functions and constraint optimization, ensuring both quality thresholds and maximal semantic differentiation. Experiments show SemDiD consistently outperforms existing methods, improving Best-of-N coverage by 1.4-5.2% across diverse tasks and accelerating RLHF training convergence by 15% while increasing accuracy by up to 2.1%.

Primary Area: Deep learning (e.g., architectures, generative models, optimization for deep networks, foundation models, LLMs)

Submission Number: 11978