RoCoFT: Efficient Finetuning of Large Language Models with Row-Column Updates

We propose Row-Column Fine-Tuning (Ro-CoFT), a parameter-efficient finetuning method for large language models based on updating only a few rows and columns of the weight matrices in transformers. Through extensive experiments with medium size language models like RoBERTa and DeBERTa, and large language models (LLMs) liken Bloom-7B, Llama2-7B and Llama2-13B, we show that our method gives comparable accuracies to the state-of-the-art Parameter-Efficient Finetuning methods while also being more memory and computation-efficient. We also study the reason behind the effectiveness of our method with tools from Neural Tangent Kernel (NTK) theory. We empirically demonstrate that our kernel, constructed using a restricted set of row and column parameters, is numerically close to the full-parameter kernel and gives comparable classification performance. Ablation studies are conducted to investigate the impact of different algorithmic choices, including the robustness of RoCoFT to any selection of rows and columns, as well as the optimal rank for the effective implementation of our method. * This work was done during an internship at Nokia Bell Labs.