Summary of Change Is the Only Constant: Dynamic Llm Slicing Based on Layer Redundancy, by Razvan-gabriel Dumitru et al.
Change Is the Only Constant: Dynamic LLM Slicing based on Layer Redundancy
by Razvan-Gabriel Dumitru, Paul-Ioan Clotan, Vikas Yadav, Darius Peteleaza, Mihai Surdeanu
First submitted to arxiv on: 5 Nov 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
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 paper introduces a novel approach to model compression in Large Language Models (LLMs) through dynamic layer-specific pruning. This method builds upon SliceGPT and leverages the Layer Redundancy (LR) score, which measures how much each layer changes its input. The LR score is used to prune individual layers based on redundancy, aiming for a fixed average pruned percentage across all layers. Experiments were conducted using LLMs like Llama3-8B and Mistral-7B on multiple datasets, evaluating different slicing bases and percentages to determine optimal configurations balancing efficiency and performance. The results show that the dynamic slicing approach maintains or even enhances model performance compared to the baseline established by constant slicing methods. For example, performance improvements of up to 5% were observed in several settings, with a perplexity decrease of as much as 7% across multiple benchmarks, validating the effectiveness of the method. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper introduces a new way to make large language models smaller and more efficient. It uses a special score called Layer Redundancy (LR) to figure out which parts of the model can be removed without hurting its performance. The LR score is used to remove layers one by one, making sure that the overall removal amount stays the same for all layers. The team tested this method on different models and datasets and found that it often performs better or just as well as older methods. For example, they saw a 5% boost in performance in some cases and a 7% decrease in mistakes made across multiple tests. |
Keywords
» Artificial intelligence » Model compression » Perplexity » Pruning
