BLoB: Bayesian Low-Rank Adaptation by Backpropagation for Large Language Models

TL;DR

BLoB introduces a Bayesian low-rank adaptation method that jointly fine-tunes mean and covariance of LLM parameters during training, improving uncertainty estimation and generalization for domain-specific tasks.

Contribution

It proposes a novel Bayesian adaptation algorithm that updates both mean and covariance during fine-tuning, unlike previous post-training Bayesian methods.

Findings

Enhanced uncertainty estimation on in-distribution data.

Improved generalization to out-of-distribution data.

Effective joint adjustment of mean and covariance during training.

Abstract

Large Language Models (LLMs) often suffer from overconfidence during inference, particularly when adapted to downstream domain-specific tasks with limited data. Previous work addresses this issue by employing approximate Bayesian estimation after the LLMs are trained, enabling them to quantify uncertainty. However, such post-training approaches' performance is severely limited by the parameters learned during training. In this paper, we go beyond post-training Bayesianization and propose Bayesian Low-Rank Adaptation by Backpropagation (BLoB), an algorithm that continuously and jointly adjusts both the mean and covariance of LLM parameters throughout the whole fine-tuning process. Our empirical results verify the effectiveness of BLoB in terms of generalization and uncertainty estimation, when evaluated on both in-distribution and out-of-distribution data.