Preprint
Jul 2026
RSRA: Training-Free Probing of Representation Sensitivity for Efficient LoRA Rank Allocation
RSRA is proposed, a training-free rank allocator that estimates where adaptation capacity is most needed through forward-only representation sensitivity probing on a small calibration set and achieves the highest average performance in all three reported model-benchmark settings and a 1.93x speedup in allocation time o...
Jia-Qi Liu, Hai-Dong Kang, Qi-Hui Zhao et al.
· 1 citation