Summary of Attention Score Is Not All You Need For Token Importance Indicator in Kv Cache Reduction: Value Also Matters, by Zhiyu Guo et al.
Attention Score is not All You Need for Token Importance Indicator in KV Cache Reduction: Value Also Matters
by Zhiyu Guo, Hidetaka Kamigaito, Taro Watanabe
First submitted to arxiv on: 18 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 proposed Value-Aware Token Pruning (VATP) method improves the scaling of large language models (LLMs) for various tasks like book summarization. By combining attention scores with value vector norms, VATP outperforms previous methods in over 12 LongBench tasks. The approach tackles the memory cost limitation of LLMs’ Key and Value (KV) cache in attention, enabling more practical applications. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models can do lots of new things, like summarize books! But there’s a problem: they use too much memory. Scientists have been trying to fix this by making some parts smaller. They thought that if they looked at how important each piece of information was, they could make it work better. But now, we know that’s not the whole story. We found out that some pieces of information are way more important than others. So, we came up with a new idea: using both how important something is and how big it is to decide what to keep or get rid of. And it works! |
Keywords
» Artificial intelligence » Attention » Pruning » Summarization » Token
