Summary of A Training-free Sub-quadratic Cost Transformer Model Serving Framework with Hierarchically Pruned Attention, by Heejun Lee et al.
A Training-free Sub-quadratic Cost Transformer Model Serving Framework With Hierarchically Pruned Attention
by Heejun Lee, Geon Park, Youngwan Lee, Jaduk Suh, Jina Kim, Wonyoung Jeong, Bumsik Kim, Hyemin Lee, Myeongjae Jeon, Sung Ju Hwang
First submitted to arxiv on: 14 Jun 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Computer Vision and Pattern Recognition (cs.CV); Distributed, Parallel, and Cluster Computing (cs.DC); 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 This abstract presents a novel approach for improving the context length in large language models (LLMs) by reducing the time and space complexities of the attention mechanism. The authors propose Hierarchically Pruned Attention (HiP), which achieves an O(T log T) time complexity and O(T) space complexity, making it more practical for real-world applications. HiP utilizes a tree-search-like algorithm that estimates top-k key tokens for a given query on the fly, based on attention locality patterns observed in pretrained LLMs. Additionally, the authors optimize GPU memory usage by implementing KV cache offloading, which stores only O(log T) tokens on the GPU while maintaining similar decoding throughput. Experiments show that HiP reduces prefill and decoding latencies, as well as memory usage, while maintaining high-quality generation with minimal degradation. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about making big language models better by allowing them to understand longer pieces of text. The authors want to make it possible for these models to process millions of words at a time, which would be useful for many applications like chatbots and language translation. They propose a new way of doing attention in these models, called HiP, that is faster and uses less memory than current methods. This means that computers can handle the processing requirements without slowing down or running out of space. The authors tested their method and found that it works well and is much faster than existing approaches. |
Keywords
* Artificial intelligence * Attention * Context length * Translation
