Rethinking Compression: Reduced Order Modelling of Latent Features in Large Language Models
This addresses the problem of high computational demands for LLM compression, particularly on consumer hardware, though it appears incremental as it builds on existing compression techniques.
The paper tackles the impracticality of compressing large language models (LLMs) with conventional methods by introducing a reduced order modelling approach that uses low-rank decomposition and re-parameterization, enabling compression of billion-scale models without GPUs and outperforming state-of-the-art structured pruning.
Due to the substantial scale of Large Language Models (LLMs), the direct application of conventional compression methodologies proves impractical. The computational demands associated with even minimal gradient updates present challenges, particularly on consumer-grade hardware. This paper introduces an innovative approach for the parametric and practical compression of LLMs based on reduced order modelling, which entails low-rank decomposition within the feature space and re-parameterization in the weight space. Notably, this compression technique operates in a layer-wise manner, obviating the need for a GPU device and enabling the compression of billion-scale models within stringent constraints of both memory and time. Our method represents a significant advancement in model compression by leveraging matrix decomposition, demonstrating superior efficacy compared to the prevailing state-of-the-art structured pruning method.