Summary of Ada-kv: Optimizing Kv Cache Eviction by Adaptive Budget Allocation For Efficient Llm Inference, By Yuan Feng and Junlin Lv and Yukun Cao and Xike Xie and S. Kevin Zhou
Ada-KV: Optimizing KV Cache Eviction by Adaptive Budget Allocation for Efficient LLM Inference
by Yuan Feng, Junlin Lv, Yukun Cao, Xike Xie, S. Kevin Zhou
First submitted to arxiv on: 16 Jul 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
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 paper investigates the efficiency challenges faced by Large Language Models in various domains, particularly with regards to their Key-Value (KV) cache required for long-sequence inference. The authors propose a head-wise adaptive budget allocation strategy called Ada-KV, which enables seamless integration with prior cache eviction methods while offering quality improvements over existing approaches. This is achieved through a theoretical loss upper bound established between pre- and post-eviction attention output, guiding the optimization of adaptive budget allocation. The paper presents extensive evaluations on 13 datasets from Ruler and 16 datasets from LongBench, conducted under both question-aware and question-agnostic scenarios. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper tries to solve a big problem with language models that need a lot of memory to work well. Right now, people are trying to make these models use less memory while still keeping the good results they get. The problem is that each part of the model needs different amounts of memory, and current solutions don’t take this into account. This paper creates a new way to decide how much memory each part gets, which makes the model work better. They tested it on many datasets and showed that it really does improve things. |
Keywords
» Artificial intelligence » Attention » Inference » Optimization
