LoRA+: Efficient Low Rank Adaptation of Large Models
This addresses an efficiency and performance bottleneck in finetuning large language models, offering a simple, incremental improvement over the widely used LoRA method.
The paper tackled the suboptimal finetuning of large models using Low Rank Adaptation (LoRA) by identifying that using the same learning rate for adapter matrices A and B leads to inefficient feature learning in wide networks. The result was LoRA+, which sets different learning rates for A and B, achieving 1-2% performance improvements and up to 2X speedup in finetuning at the same computational cost as LoRA.
In this paper, we show that Low Rank Adaptation (LoRA) as originally introduced in Hu et al. (2021) leads to suboptimal finetuning of models with large width (embedding dimension). This is due to the fact that adapter matrices A and B in LoRA are updated with the same learning rate. Using scaling arguments for large width networks, we demonstrate that using the same learning rate for A and B does not allow efficient feature learning. We then show that this suboptimality of LoRA can be corrected simply by setting different learning rates for the LoRA adapter matrices A and B with a well-chosen ratio. We call this proposed algorithm LoRA$+$. In our extensive experiments, LoRA$+$ improves performance (1-2 $\%$ improvements) and finetuning speed (up to $\sim$ 2X SpeedUp), at the same computational cost as LoRA.