Summary of Gateau: Selecting Influential Samples For Long Context Alignment, by Shuzheng Si et al.
GATEAU: Selecting Influential Samples for Long Context Alignment
by Shuzheng Si, Haozhe Zhao, Gang Chen, Yunshui Li, Kangyang Luo, Chuancheng Lv, Kaikai An, Fanchao Qi, Baobao Chang, Maosong Sun
First submitted to arxiv on: 21 Oct 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 proposes a novel framework called GATEAU for aligning large language models with extremely long contexts, which has yet to be fully investigated. The current approach synthesizes long instruction-following samples by scaling up the available data volume, but lacks a well-defined strategy for ensuring data quality. This may introduce low-quality samples and restrict model performance. The authors identify influential samples enriched with long-range dependency relations, measuring long-range dependencies from two aspects: generating target responses and understanding long inputs. Experiments show that GATEAU effectively identifies influential samples, leading to better instruction-following and long-context understanding capabilities. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about helping computers understand very long instructions. Right now, it’s hard for computers to learn from these types of instructions because there isn’t a good way to prepare the data. The authors are trying to solve this problem by finding important samples that have special relationships between different parts of the instruction. They want to see if using these special samples can help computers understand long instructions better. |
