Dist2ill: Distributional Distillation for One-Pass Uncertainty Estimation in Large Language Models

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Zhao, Yicong, Tsang, King Yeung, Vejendla, Harshil, Shi, Haizhou, Li, Zhuohang, Hua, Zhigang, Xu, Qi, Zhang, Tunyu, Wang, Yi, Han, Ligong, Malin, Bradley A., Wang, Hao
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866918328246730752
author Zhao, Yicong
Tsang, King Yeung
Vejendla, Harshil
Shi, Haizhou
Li, Zhuohang
Hua, Zhigang
Xu, Qi
Zhang, Tunyu
Wang, Yi
Han, Ligong
Malin, Bradley A.
Wang, Hao
author_facet Zhao, Yicong
Tsang, King Yeung
Vejendla, Harshil
Shi, Haizhou
Li, Zhuohang
Hua, Zhigang
Xu, Qi
Zhang, Tunyu
Wang, Yi
Han, Ligong
Malin, Bradley A.
Wang, Hao
contents Large Language Models (LLMs) often exhibit misalignment between the quality of their generated responses and the confidence estimates they assign to them. Bayesian treatments, such as marginalizing over a reliable weight posterior or over the space of reasoning traces, provide an effective remedy, but incur substantial computational overhead due to repeated sampling at test time. To enable accurate uncertainty estimation in a single forward pass, we propose a novel distributional distillation framework (Dist2ill) that trains an LLM to produce multiple diverse reasoning paths within one inference pass, while using a lightweight parametric module to approximate empirical confidence scores derived from the sampling distribution. Extensive experiments demonstrate that Dist2ill preserves reasoning diversity and achieves state-of-the-art uncertainty estimation, substantially improving Expected Calibration Error (ECE) and Negative Log-Likelihood (NLL), while remaining computationally efficient.
format Preprint
id arxiv_https___arxiv_org_abs_2505_11731
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Dist2ill: Distributional Distillation for One-Pass Uncertainty Estimation in Large Language Models
Zhao, Yicong
Tsang, King Yeung
Vejendla, Harshil
Shi, Haizhou
Li, Zhuohang
Hua, Zhigang
Xu, Qi
Zhang, Tunyu
Wang, Yi
Han, Ligong
Malin, Bradley A.
Wang, Hao
Machine Learning
Artificial Intelligence
Computation and Language
Large Language Models (LLMs) often exhibit misalignment between the quality of their generated responses and the confidence estimates they assign to them. Bayesian treatments, such as marginalizing over a reliable weight posterior or over the space of reasoning traces, provide an effective remedy, but incur substantial computational overhead due to repeated sampling at test time. To enable accurate uncertainty estimation in a single forward pass, we propose a novel distributional distillation framework (Dist2ill) that trains an LLM to produce multiple diverse reasoning paths within one inference pass, while using a lightweight parametric module to approximate empirical confidence scores derived from the sampling distribution. Extensive experiments demonstrate that Dist2ill preserves reasoning diversity and achieves state-of-the-art uncertainty estimation, substantially improving Expected Calibration Error (ECE) and Negative Log-Likelihood (NLL), while remaining computationally efficient.
title Dist2ill: Distributional Distillation for One-Pass Uncertainty Estimation in Large Language Models
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2505.11731