Summary of Post Training Quantization Of Large Language Models with Microscaling Formats, by Sayeh Sharify et al.
Post Training Quantization of Large Language Models with Microscaling Formats
by Sayeh Sharify, Utkarsh Saxena, Zifei Xu, Wanzin Yazar, Ilya Soloveychik, Xin Wang
First submitted to arxiv on: 12 May 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 impressive performance in complex language modeling tasks, but at a significant computational and storage cost. This paper investigates the potential of quantization to alleviate these challenges. We systematically examine the combination of three post-training techniques: SmoothQuant, AWQ, and GPTQ. Our analysis reveals the interactions and implications of applying these methods for advancing LLM quantization. By enabling quantization in microscaling (MX) formats, we expand the applicability of PTQ algorithms beyond their original fixed-point format targets. We demonstrate that combining different PTQ methods allows us to quantize models to 4-bit weights and 8-bit activations using the MXINT format with minimal accuracy loss compared to the uncompressed baseline. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large Language Models (LLMs) are really good at understanding language, but they use a lot of computer power. This paper looks into ways to make them more efficient. We test three special techniques: SmoothQuant, AWQ, and GPTQ. We see how these methods work together and what this means for making LLMs better. By changing the way we store information, we can make LLMs use less computer power while still being very accurate. |
Keywords
» Artificial intelligence » Quantization
