From Isolation to Alignment: Unified LoRA for Efficient Multi-Task Learning
This work reveals a paradox where a simplified, router-free multi-head model with high inter-head redundancy outperforms complex, diversity-driven baselines and proposes Align-LoRA, a unified and efficient framework that shifts the focus from architectural isolation to representation alignment.
Jinda Liu, Bo Cheng, Yi Chang et al.
· 1 citation