AI Dynamics

Global AI News Aggregator

About

LoRA+: Optimizing Learning Rates for Matrix B in Fine-tuning

5) LoRA+ – In LoRA, both matrices A and B are updated with the same learning rate.
– Authors of LoRA+ found that setting a higher learning rate for matrix B results in better convergence. Check this

→ View original post on X — @akshay_pachaar