Astra: Activation-Space Tail-Eigenvector Low-Rank Adaptation of Large Language Models

TL;DR

Astra introduces a novel PEFT method that exploits activation tail eigenvectors to create efficient, task-adaptive low-rank adapters, leading to faster convergence and better performance than existing methods.

Contribution

Astra leverages tail eigenvectors of activation subspaces for low-rank adaptation, enhancing fine-tuning efficiency and effectiveness over prior PEFT approaches.

Findings

Outperforms existing PEFT methods on 16 benchmarks.

Achieves faster convergence and better downstream performance.

Surpasses full fine-tuning in certain tasks.

Abstract

Parameter-Efficient Fine-Tuning (PEFT) methods, especially LoRA, are widely used for adapting pre-trained models to downstream tasks due to their computational and storage efficiency. However, in the context of LoRA and its variants, the potential of activation subspaces corresponding to tail eigenvectors remains substantially under-exploited, which may lead to suboptimal fine-tuning performance. In this work, we propose Astra (Activation-Space Tail-Eigenvector Low-Rank Adaptation), a novel PEFT method that leverages the tail eigenvectors of the model output activations-estimated from a small task-specific calibration set-to construct task-adaptive low-rank adapters. By constraining updates to the subspace spanned by these tail eigenvectors, Astra achieves faster convergence and improved downstream performance with a significantly reduced parameter budget. Extensive experiments across natural language understanding (NLU) and natural language generation (NLG) tasks demonstrate that Astra consistently outperforms existing PEFT baselines across 16 benchmarks and even surpasses full fine-tuning (FFT) in certain scenarios.