通过最小化KL散度，实现约束解码中概率分布的精准保持。
(G)I-DLE: Generative Inference via Distribution-preserving Logit Exclusion with KL Divergence Minimization for Constrained Decoding
- 用KL散度最小化重构合法词元概率，避免传统置负无穷带来的分布扭曲
- 在1.5B到14B的Qwen2.5模型上提升平均评分并显著降低输出质量方差
- 适合需要高稳定性和语言合规性的生成任务，如韩语内容生成
我们提出(G)I-DLE，一种新型约束解码方法，通过最小化KL散度来保留自回归语言模型的固有条件概率分布，同时排除不良词元。与传统方法将禁用词元的logits设为$-\infty$不同，该方法会重新归一化允许词元的概率以最小化分布失真。我们在专用于评估韩语流利度、逻辑推理和文化适宜性的K2-Eval数据集上验证了该方法。实验结果表明，在1.5B至14B的Qwen2.5模型上，G-IDLE不仅提升了平均评估得分，还显著降低了输出质量的方差。
原文摘要 · Abstract (English)
We propose (G)I-DLE, a new approach to constrained decoding that leverages KL divergence minimization to preserve the intrinsic conditional probability distribution of autoregressive language models while excluding undesirable tokens. Unlike conventional methods that naively set banned tokens' logits to $-\infty$, which can distort the conversion from raw logits to posterior probabilities and increase output variance, (G)I-DLE re-normalizes the allowed token probabilities to minimize such distortion. We validate our method on the K2-Eval dataset, specifically designed to assess Korean language fluency, logical reasoning, and cultural appropriateness. Experimental results on Qwen2.5 models (ranging from 1.5B to 14B) demonstrate that G-IDLE not only boosts mean evaluation scores but also substantially reduces the variance of output quality.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。