AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

BoRA: Towards More Expressive Low-Rank Adaptation with Block Diversity

Shiwei Li, Xiandi Luo, Haozhao Wang, Xing Tang, Ziqiang Cui, Dugang Liu, Yuhua Li, Yichen Li, Xiuqiang He, Ruixuan Li

ICLR 2026 · Architectures and efficiency › Pruning and efficient tuning · approach: LoRA variants and theory · 6 citations (OpenAlex)

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 changedWhereMeasuredEffect
Replace LoRA r=8 with BoRA r=8, b=8LoRA · GLUE avg, RoBERTa-Baseaccuracy (%): 82.55 -> 83.65Improved
Replace LoRA r=8 with BoRA r=8, b=16LoRA · GLUE avg, RoBERTa-Largeaccuracy (%): 84.38 -> 86.51Improved
Replace LoRA r=8 with BoRA r=8, b=16LoRA · Math10K avg, LLaMA-3-8Baccuracy (%): 68.19 -> 72.24Improved
Replace LoRA r=8 with BoRA r=8, b=16LoRA · Commonsense170K avg, LLaMA-3-8Baccuracy (%): 85.45 -> 86.62Improved
Use BoRA r=8 b=16 instead of 4x larger LoRA r=32LoRA r=32 · Math10K avg, Qwen2.5-14Baccuracy (%): 80.15 -> 80.60Improved
Use BoRA r=8 b=16 instead of LoRA r=32LoRA r=32 · Commonsense170K avg, LLaMA-3-8Baccuracy (%): 86.72 -> 86.62No clear effect
Use MELoRA (zeroed off-diagonal blocks) instead of LoRALoRA r=8 · GLUE avg, RoBERTa-Baseaccuracy (%): 82.55 -> 81.74Hurt
Increase blocks from 8 to 16 (ablation)BoRA r=8 · GLUE avg, RoBERTa-Largeaccuracy (%): 85.83 -> 86.51Improved
Add exponential function to sigma parametrization (ablation)BoRA · Math10KaccuracyImproved
Add mean-absolute-value normalization to sigma (ablation)BoRA · Math10KaccuracyImproved
Increase block number b at fixed rank (ablation)BoRA · Math10K, LLaMA-3-8BaccuracyMixed
Apply BoRA vs LoRA at different tuned modules (Q, QV, QKVUD, QKVOGUD) (ablation)LoRA · Math10K, LLaMA-3-8BaccuracyImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy