Bayesian-LoRA: Probabilistic Low-Rank Adaptation of Large Language Models
About
Large Language Models usually put more emphasis on accuracy and therefore, will guess even when not certain about the prediction, which is especially severe when fine-tuned on small datasets due to the inherent tendency toward miscalibration. In this work, we introduce Bayesian-LoRA, which reformulates the deterministic LoRA update as a probabilistic low-rank representation inspired by Sparse Gaussian Processes. We identify a structural isomorphism between LoRA's factorization and Kronecker-factored SGP posteriors, and show that LoRA emerges as a limiting case when posterior uncertainty collapses. We conduct extensive experiments on various LLM architectures across commonsense reasoning benchmarks. With only approximately 0.42M additional parameters and ${\approx}1.2{\times}$ training cost relative to standard LoRA, Bayesian-LoRA significantly improves calibration across models up to 30B, achieving up to 84% ECE reduction and 76% NLL reduction while maintaining competitive accuracy for both in-distribution and out-of-distribution (OoD) evaluations.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Commonsense Reasoning | ARC Challenge | Accuracy68.9 | 132 | |
| Commonsense Reasoning | ARC-E | Accuracy85.91 | 62 | |
| Commonsense Reasoning | WG-S | Accuracy70.9 | 18 | |
| Commonsense Reasoning | WG-M | Accuracy74.3 | 18 | |
| Commonsense Reasoning | BoolQ | Accuracy86.1 | 18 | |
| Mathematical Reasoning | MATH | CoT NLL0.513 | 11 | |
| Commonsense Reasoning | OpenBookQA | ACC81.6 | 9 | |
| Question Answering | OBQA in-distribution (test) | Accuracy81.6 | 9 | |
| Question Answering | ARC-C Small Shift (test) | Accuracy69.5 | 9 | |
| Question Answering | ARC-E Small Shift (test) | Accuracy78.9 | 9 |