Summary of Quest: Query-aware Sparsity For Efficient Long-context Llm Inference, by Jiaming Tang et al.
Quest: Query-Aware Sparsity for Efficient Long-Context LLM Inference
by Jiaming Tang, Yilong Zhao, Kan Zhu, Guangxuan Xiao, Baris Kasikci, Song Han
First submitted to arxiv on: 16 Jun 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 paper proposes Quest, a query-aware KV cache selection algorithm to speed up self-attention in long-context large language models (LLMs). The authors observe that the criticality of tokens depends on the query and only a small portion of tokens dominates attention outcomes. To address this challenge, Quest estimates the criticality of a given page using Query vectors and loads only the Top-K critical KV cache pages for attention, achieving up to 2.23x self-attention speedup with negligible accuracy loss. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Quest helps improve inference speed in long-context LLMs by reducing the time it takes to load large KV caches during self-attention. This is especially important as models with longer context windows become more common. The algorithm works by identifying critical tokens and loading only the necessary pages, allowing for faster computation without sacrificing accuracy. |
Keywords
* Artificial intelligence * Attention * Inference * Self attention
