Summary of Dijiang: Efficient Large Language Models Through Compact Kernelization, by Hanting Chen et al.
DiJiang: Efficient Large Language Models through Compact Kernelization
by Hanting Chen, Zhicheng Liu, Xutao Wang, Yuchuan Tian, Yunhe Wang
First submitted to arxiv on: 29 Mar 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: 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 proposed Frequency Domain Kernelization approach, DiJiang, transforms a pre-trained vanilla Transformer into a linear complexity model with minimal training costs. This is achieved by employing a weighted Quasi-Monte Carlo method for sampling and Discrete Cosine Transform (DCT) operations to reduce computational complexity. The approach enables comparable performance to the original Transformer on various benchmarks while requiring significantly reduced training costs and faster inference speeds. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper introduces DiJiang, a way to make Transformers less computationally heavy. They do this by changing the attention mechanism in the Transformer model without needing to retrain the whole model. This makes it much faster and cheaper to train large language models. The new approach uses a special kind of sampling called Quasi-Monte Carlo and some mathematical tricks with DCT. The results show that DiJiang performs just as well as the original Transformer, but is much faster and cheaper. |
Keywords
» Artificial intelligence » Attention » Inference » Transformer
