Abstract
Fisher information is a fundamental concept for characterizing the sensitivity of parameters in neural networks. However, leveraging the full observed Fisher information is too expensive for large models, so most methods rely on simple diagonal approximations. While efficient, this approach ignores parameter correlations, often resulting in reduced performance on downstream tasks.
In this work, we propose Generalized Fisher-Weighted SVD (GFWSVD), a post-training LLM compression technique that accounts for both diagonal and off-diagonal elements of the Fisher information matrix, providing a more accurate reflection of parameter importance. To make the method tractable, we introduce a scalable adaptation of the Kronecker-factored approximation algorithm for the observed Fisher information.
We demonstrate the effectiveness of our method on LLM compression, showing improvements over existing compression baselines. For instance, at a 20x compression rate on the MMLU benchmark, our method outperforms FWSVD, which is based on a diagonal approximation of the Fisher information, by 5 percent, SVD-LLM by 3 percent, and ASVD by 6 percent.
Blogger's Review: The GFWSVD method significantly enhances the model compression effectiveness by considering the full scope of Fisher information, particularly showcasing its strong potential in large language model applications. This method offers new insights for future research and applications.