Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2501.14713 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916591724134400 |
|---|---|
| author | Smith, James Seale Lin, Chi-Heng Tuli, Shikhar Jeelani, Haris Gao, Shangqian Shen, Yilin Jin, Hongxia Hsu, Yen-Chang |
| author_facet | Smith, James Seale Lin, Chi-Heng Tuli, Shikhar Jeelani, Haris Gao, Shangqian Shen, Yilin Jin, Hongxia Hsu, Yen-Chang |
| contents | The rapid proliferation of large language models (LLMs) in natural language processing (NLP) has created a critical need for techniques that enable efficient deployment on memory-constrained devices without compromising performance. We present a method to prune LLMs that selectively prunes model blocks based on an importance score and replaces them with a low-parameter replacement strategy. Specifically, we propose a principled metric to replace each pruned block using a weight-sharing mechanism that leverages unpruned counterparts from the model and block-specific low-rank adapters. Furthermore, we facilitate the learning of these replacement blocks with output feature normalization and an adapter initialization scheme built on low-rank SVD reconstructions. Empirical evaluations demonstrate substantial performance gains over existing methods, achieving state-of-the-art performance on 5/6 benchmarks for a compression rate of 30% and 6/6 benchmarks for a compression rate of 40%. We also demonstrate that our approach can extend smaller models, boosting performance on 6/6 benchmarks using only ~0.3% tokens of extended training with minimal additional parameter costs. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2501_14713 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | FlexiGPT: Pruning and Extending Large Language Models with Low-Rank Weight Sharing Smith, James Seale Lin, Chi-Heng Tuli, Shikhar Jeelani, Haris Gao, Shangqian Shen, Yilin Jin, Hongxia Hsu, Yen-Chang Computation and Language Machine Learning The rapid proliferation of large language models (LLMs) in natural language processing (NLP) has created a critical need for techniques that enable efficient deployment on memory-constrained devices without compromising performance. We present a method to prune LLMs that selectively prunes model blocks based on an importance score and replaces them with a low-parameter replacement strategy. Specifically, we propose a principled metric to replace each pruned block using a weight-sharing mechanism that leverages unpruned counterparts from the model and block-specific low-rank adapters. Furthermore, we facilitate the learning of these replacement blocks with output feature normalization and an adapter initialization scheme built on low-rank SVD reconstructions. Empirical evaluations demonstrate substantial performance gains over existing methods, achieving state-of-the-art performance on 5/6 benchmarks for a compression rate of 30% and 6/6 benchmarks for a compression rate of 40%. We also demonstrate that our approach can extend smaller models, boosting performance on 6/6 benchmarks using only ~0.3% tokens of extended training with minimal additional parameter costs. |
| title | FlexiGPT: Pruning and Extending Large Language Models with Low-Rank Weight Sharing |
| topic | Computation and Language Machine Learning |
| url | https://arxiv.org/abs/2501.14713 |