Summary of Mix-ln: Unleashing the Power Of Deeper Layers by Combining Pre-ln and Post-ln, By Pengxiang Li et al.
Mix-LN: Unleashing the Power of Deeper Layers by Combining Pre-LN and Post-LN
by Pengxiang Li, Lu Yin, Shiwei Liu
First submitted to arxiv on: 18 Dec 2024
Categories
- Main: Machine Learning (cs.LG)
- 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 Large Language Models (LLMs) have achieved remarkable success, but recent findings show that their deeper layers often contribute minimally and can be pruned without affecting performance. This paper identifies a training shortfall caused by the widespread use of Pre-Layer Normalization (Pre-LN) in models like GPT and LLaMA, which leads to diminished gradient norms in deeper layers. To address this, the authors introduce Mix-LN, a novel normalization technique that combines strengths of Pre-LN and Post-LN. Extensive experiments with various model sizes demonstrate that Mix-LN consistently outperforms both Pre-LN and Post-LN, promoting more balanced gradients throughout the network, enhancing overall quality of LLM pre-training. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large Language Models have been super successful, but some parts don’t work as well as they should. This paper figures out why this is happening because of a common way models are trained called Pre-Layer Normalization (Pre-LN). The authors then create a new way to train called Mix-LN that combines good things from old ways and makes models better. |
Keywords
» Artificial intelligence » Gpt » Llama
