Summary of Interactions Across Blocks in Post-training Quantization Of Large Language Models, by Khasmamad Shabanovi et al.
Interactions Across Blocks in Post-Training Quantization of Large Language Models
by Khasmamad Shabanovi, Lukas Wiest, Vladimir Golkov, Daniel Cremers, Thomas Pfeil
First submitted to arxiv on: 6 Nov 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); 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 The paper investigates the effects of simplifying assumptions on post-training quantization of large language models. Specifically, it examines two multi-block fine-tuning strategies and compares them to the baseline of fine-tuning single transformer blocks. The first strategy captures correlations across blocks by jointly optimizing multiple quantized blocks, while the second incorporates knowledge of subsequent blocks by minimizing error in downstream pre-activations. The study finds that the effectiveness of these methods depends on the specific network model, with significant benefits for some models but no impact for others. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper looks at how to make neural networks use less energy and calculations. It usually does this by changing individual parts of the network, like layers, to use fewer numbers. This helps reduce errors in what the network thinks before it acts. The researchers wanted to see if these simplifications work when applied to big language models that can understand lots of text. They came up with two new ways to adjust these big models and tested them against just adjusting one part at a time. They found out that which method works best depends on what kind of model you’re using. |
Keywords
» Artificial intelligence » Fine tuning » Quantization » Transformer
