analiza aiqbench
This research introduces DALorRA, a variational Bayesian framework that improves LLM calibration by applying stochastic masking to LoRA rank dimensions rather than dense parameters. By treating low-rank components as an ensemble, the method effectively reduces overconfidence during task-specific fine-tuning without sacrificing reasoning accuracy. The approach offers a principled path toward more trustworthy deployment of large language models through efficient uncertainty quantification.
Bayesian Sparse Low-Rank Adaptation for Large Language Model Uncertainty Estimation
arXiv:2607.02182v2 Announce Type: replace-cross Abstract: Large language models (LLMs) exhibit remarkable reasoning capabilities, but their task-specific fine-tuning is notoriously plagued by overconfidence, severely hindering trustworthy deployment. We propose Data-Adaptive Lower-Rank Adaptation (DALorRA), a simple and effective variational Bayesian sparse framework that shifts the paradigm of uncertainty quantification from the dense parameter space to the lightweight rank level of low-rank adaptation (LoRA). With the insight that LoRA essentially aggregates multiple rank-one components that may provide superfluous model capacity, DALorRA imposes stochastic masking on rank dimensions, enabling Bayesian regularization of model capacity during training and ensemble-like calibration during inference. Extensive experiments demonstrate DALorRA's excellent calibration of LLMs without compromising reasoning accuracy.
Abstract:Large language models (LLMs) exhibit remarkable reasoning capabilities, but their task-specific fine-tuning is notoriously plagued by overconfidence, severely hindering trustworthy deployment. We propose Data-Adaptive Lower-Rank Adaptation (DALorRA), a simple and effective variational Bayesian sparse framework that shifts the paradigm of uncertainty quantification from the dense parameter space to the lightweight rank level of low-rank adaptation (LoRA). With the insight that LoRA essentially aggregates multiple rank-one components that may provide superfluous model capacity, DALorRA imposes stochastic masking on rank dimensions, enabling Bayesian regularization of model capacity during training and ensemble-like calibration during inference. Extensive experiments demonstrate DALorRA's excellent calibration of LLMs without compromising reasoning accuracy.
To appear in EMNLP 2026. 17 pages, 7 figures, 8 tables
Machine Learning (cs.LG); Computation and Language (cs.CL)
arXiv:2607.02182 [cs.LG]
(or arXiv:2607.02182v2 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2607.02182 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 211 cuvinte
Subiecte conexe
Bayesian-LoRA: Probabilistic Low-Rank Adaptation of Large Language Models
arXiv · 28 iul. 2026
The Continual Constitutional Convention: How language models revise the documents that govern them
AI Alignment Forum · 13 oct. 2026
Amazon is investing in the Lean Focused Research Organization
Amazon Science · 26 iul. 2026
PIRAMID: Progress and Plans
LessWrong · 27 iul. 2026