Summary of Pruning Large Language Models with Semi-structural Adaptive Sparse Training, by Weiyu Huang et al.
Pruning Large Language Models with Semi-Structural Adaptive Sparse Training
by Weiyu Huang, Yuezhou Hu, Guohao Jian, Jun Zhu, Jianfei Chen
First submitted to arxiv on: 30 Jul 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The proposed Adaptive Sparse Trainer (AST) is a novel framework that efficiently retrains Large Language Models (LLMs) to learn optimal masks during the weight update process, without incurring additional computational overhead. By incorporating knowledge distillation and well-initialized parameters, AST achieves state-of-the-art performance with minimal training cost. Specifically, when applied to the LLaMA2-7B model, AST reduces the perplexity and zero-shot accuracy gap between dense and 2:4 semi-structured sparse models to 0.6 and 1.16%, respectively, utilizing less than 0.4% of the pretraining tokens and GPU hours. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The researchers developed a new way to make Large Language Models smaller without losing their ability to understand language. They created a special training method called Adaptive Sparse Trainer (AST) that helps models learn how to leave out unnecessary parts while keeping important information. This makes it possible to use these models on devices with limited memory and processing power. The new method was tested on a large model and showed great results, making it a promising approach for creating smaller language models. |
Keywords
» Artificial intelligence » Knowledge distillation » Perplexity » Pretraining » Zero shot
