Summary of Direct Quantized Training Of Language Models with Stochastic Rounding, by Kaiyan Zhao et al.
Direct Quantized Training of Language Models with Stochastic Rounding
by Kaiyan Zhao, Tsuguchika Tabaru, Kenichi Kobayashi, Takumi Honda, Masafumi Yamazaki, Yoshimasa Tsuruoka
First submitted to arxiv on: 6 Dec 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 A novel approach is proposed to reduce memory footprint during training of Large Language Models (LLMs). Recent quantized LLMs have demonstrated significant memory reduction at deployment time, but training these models still requires substantial memory. The authors explore direct updates to quantized low-precision weight matrices without relying on straight-through estimation, aiming to minimize information loss and save memory usage during training. Experiments on LLaMA-structured models show that training with ternary weights is feasible, and increasing the bit width to 8 bits results in only a 5% loss degradation compared to BitNet b1.58. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large Language Models are super smart computers that can understand and generate human-like text. Right now, training these models takes up a lot of memory space. Researchers have found ways to reduce this memory usage when the model is used in real-life applications. But what if we could train these models using even less memory? This paper explores new techniques for doing just that. The authors tested their approach on some really cool models and found that it works! They were able to train these models using ternary weights (which are like super simplified versions of regular numbers) without losing much accuracy. |
Keywords
» Artificial intelligence » Llama » Precision
