作者位次 #3/5ACL 2026Findings· Findings of ACL
Establishing a Scale for K ullback- L eibler Divergence in Language Models Across Various Settings
Ryo Kishino · Yusuke Takase · Momose Oyama · Hiroaki Yamagiwa · Hidetoshi Shimodaira
摘要
Log-likelihood vectors define a common space for comparing language models as probability distributions, enabling unified comparisons across heterogeneous settings. We extend this framework to training checkpoints and intermediate layers, and establish a consistent scale for KL divergence across...