Order-Level Attention Similarity Across Language Models: A Latent Commonality
Jinglin Liang, Jin Zhong, Shuangping Huang, Yunqing Hu, Huiyuan Zhang, Huifang Li, Lixin Fan, Hanlin Gu · 2025
Read Order-Level Attention Similarity Across Language Models: A Latent Commonality by Jinglin Liang, Jin Zhong, Shuangping Huang, Yunqing Hu, Huiyuan Zhang, Huifang Li, Lixin Fan, Hanlin Gu (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-05064.txt.
Description
Jinglin Liang and co-authors ask whether language models share hidden patterns in how they gather context. Past work looked at single models or single attention heads, so cross-model traits stayed out of view.
The team defines order-level attention, built by splitting a known attention rollout method by order. They find that this measure lines up across models at the same order, a shared trait they call a latent commonality. That shared pattern could deepen grasp of how models work and may help move knowledge from one model to another in time.
Catalog record
| CREATOR | Jinglin Liang, Jin Zhong, Shuangping Huang, Yunqing Hu, Huiyuan Zhang, Huifang Li, Lixin Fan, Hanlin Gu |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 95 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-05064