From Parameters to Performance: A Data-Driven Study on LLM Structure and Development

Suqing Wang, Zuchao Li, Luohe Shi, Bo Du, Hai Zhao, Yun Li and Qianren Wang · 2025

Read From Parameters to Performance: A Data-Driven Study on LLM Structure and Development by Suqing Wang, Zuchao Li, Luohe Shi, Bo Du, Hai Zhao, Yun Li and Qianren Wang (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-2509-18136.txt.

Description

Suqing Wang and 6 co-authors mine a large data set to link a language model's shape to its scores. Their 2025 study notes that model size and skill have grown fast, while firm facts on which shape choices matter stay scarce. The team built a set of open model shapes and their bench scores.

With that set, they test how each shape choice shifts scores, task by task. Inner-state probes back up the broad trends they find. A short past of the field, and a look at next trends, frame the sums. The set will be shared in public for folk to reuse.

Catalog record

CREATORSuqing Wang, Zuchao Li, Luohe Shi, Bo Du, Hai Zhao, Yun Li and Qianren Wang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE45 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-18136

Browse the full library · About the Tower of Babel