Summary of Less Is Ken: a Universal and Simple Non-parametric Pruning Algorithm For Large Language Models, by Michele Mastromattei et al.
Less is KEN: a Universal and Simple Non-Parametric Pruning Algorithm for Large Language Models
by Michele Mastromattei, Fabio Massimo Zanzotto
First submitted to arxiv on: 5 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computation and Language (cs.CL)
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 This paper introduces KEN, a novel neural network pruning algorithm that selectively preserves the most significant parameters while restoring others to their pre-training state. KEN is based on Kernel Density Estimation (KDE) and aims to construct optimized transformers with substantial memory savings. The algorithm achieves equal or better performance compared to original unpruned models, with a minimum parameter reduction of 25%. Extensive evaluations across seven different language models demonstrate the effectiveness of KEN, which outperforms established pruning and PEFT algorithms. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper makes it possible to make neural networks smaller without losing their ability to work well. It’s like editing a big book by deleting unnecessary words and keeping the important ones. The new algorithm, called KEN, uses special math called Kernel Density Estimation to figure out which parts of the network are most important. By getting rid of the less important parts, we can save memory and make the network work faster. The results show that KEN is very good at making networks smaller without losing their power. |
Keywords
* Artificial intelligence * Density estimation * Neural network * Pruning
