Unlocking the Global Synergies in Low-Rank Adapters
This work addresses the need for more efficient fine-tuning in NLP, but it is incremental as it builds on the established LoRA technique.
The paper tackles the problem of optimizing parameter allocation in Low-Rank Adaptation (LoRA) for fine-tuning large language models by introducing HeteroLoRA, a lightweight search algorithm that uses zero-cost proxies to allocate trainable parameters, resulting in improved model performance, such as a 1.6% accuracy gain on MRPC with a similar parameter budget.
Low-rank Adaption (LoRA) has been the de-facto parameter-efficient fine-tuning technique for large language models. We present HeteroLoRA, a light-weight search algorithm that leverages zero-cost proxies to allocate the limited LoRA trainable parameters across the model for better fine-tuned performance. In addition to the allocation for the standard LoRA-adapted models, we also demonstrate the efficacy of HeteroLoRA by performing the allocation in a more challenging search space that includes LoRA modules and LoRA-adapted shortcut connections. Experiments show that HeteroLoRA enables improvements in model performance given the same parameter budge. For example, on MRPC, we see an improvement of 1.6% in accuracy with similar training parameter budget. We will open-source our algorithm once the paper is accepted.