Summary of Llmlingua-2: Data Distillation For Efficient and Faithful Task-agnostic Prompt Compression, by Zhuoshi Pan et al.
LLMLingua-2: Data Distillation for Efficient and Faithful Task-Agnostic Prompt Compression
by Zhuoshi Pan, Qianhui Wu, Huiqiang Jiang, Menglin Xia, Xufang Luo, Jue Zhang, Qingwei Lin, Victor Rühle, Yuqing Yang, Chin-Yew Lin, H. Vicky Zhao, Lili Qiu, Dongmei Zhang
First submitted to arxiv on: 19 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 paper explores task-agnostic prompt compression techniques to enhance generalizability and efficiency. Current approaches compress prompts by removing tokens based on information entropy from a causal language model like LLaMa-7B. However, this method may be suboptimal due to limitations in capturing all essential information. The authors identify two challenges: (i) unidirectional context is not sufficient for prompt compression, and (ii) the compression objective is not aligned with the used metric. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper looks at ways to make prompts shorter and more useful. Right now, people remove words from prompts based on how important they are, using a special language model called LLaMa-7B. But this method might not be the best because it only looks at one direction of information and doesn’t fully capture what’s needed for prompt compression. |
Keywords
* Artificial intelligence * Causal language model * Language model * Llama * Prompt
