Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Bayesian-LoRA: Probabilistic Low-Rank Adaptation of Large Language Models
Moule Lin, Shuhao Guan, Andrea Patane, David Gregg, Goetz Botterweck
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add normalizing flow to the inducing posterior (L=1 vs L=0) (ablation) | Bayesian-LoRA · OBQA | accuracy: 79.0 -> 81.6 | Improved |
| add normalizing flow to the inducing posterior (L=1 vs L=0) (ablation) | Bayesian-LoRA · OBQA | NLL: 0.58 -> 0.49 | Improved |
| increase flow depth from 1 to 4 (ablation) | Bayesian-LoRA · OBQA | ECE: 5.7 -> 4.9 | Improved |
| increase flow depth from 1 to 2 (ablation) | Bayesian-LoRA · OBQA | accuracy: 81.6 -> 80.8 | Hurt |
| increase flow depth from 1 to 4 (ablation) | Bayesian-LoRA · OBQA | train time (x MAP): 1.23 -> 1.38 | Hurt |
| increase inducing dimension r=c (ablation) | Bayesian-LoRA · OBQA | ECE | Improved |
| add full probabilistic posterior over degenerate (MAP-limit) setting (ablation) | Bayesian-LoRA · OBQA | accuracy: 77.81 -> 81.60 | Improved |
| add full probabilistic posterior over degenerate (MAP-limit) setting (ablation) | Bayesian-LoRA · OBQA | ECE: 9.77 -> 5.70 | Improved |
| use Bayesian-LoRA instead of BLoB | Llama-2-7B · WinoGrande-S | accuracy: 69.07 -> 70.90 | Improved |
| use Bayesian-LoRA instead of LoRA MAP | Llama-2-7B · WinoGrande-S | ECE: 30.80 -> 4.90 | Improved |
| use Bayesian-LoRA N=2 instead of LoRA MAP | Llama-2-7B · WikiText-2 top 5% entropy tokens | ECE (test): 1.60 -> 1.26 | Improved |
| use Bayesian-LoRA instead of LoRA baseline FT | Qwen2.5-14B-Instruct · MATH | CoT-NLL: 2.165 -> 0.513 | Improved |
Closest papers by meaning
- C-LoRA: Contextual Low-Rank Adaptation for Uncertainty Estimation in Large Language Models (NeurIPS 2025 · similarity 0.94)
- Bayesian Low-rank Adaptation for Large Language Models (ICLR 2024 · similarity 0.94)
- BLoB: Bayesian Low-Rank Adaptation by Backpropagation for Large Language Models (NeurIPS 2024 · similarity 0.92)
- Make LoRA Great Again: Boosting LoRA with Adaptive Singular Values and Mixture-of-Experts Optimization Alignment (ICML 2025 · similarity 0.92)
- Training-Free Bayesianization for Low-Rank Adapters of Large Language Models (NeurIPS 2025 · similarity 0.92)
- QA-LoRA: Quantization-Aware Low-Rank Adaptation of Large Language Models (ICLR 2024 · similarity 0.92)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
