Embedding language models in probability space via log-likelihood vectors
-
Mapping 1,000+ Language Models via the Log-Likelihood Vector
Momose Oyama, Hiroaki Yamagiwa, Yusuke Takase, Hidetoshi Shimodaira
arXiv:2502.16173 | accepted to ACL 2025 main | Data & Code | Examples -
Likelihood Variance as Text Importance for Resampling Texts to Map Language Models
Momose Oyama, Ryo Kishino, Hiroaki Yamagiwa, Hidetoshi Shimodaira
arXiv:2505.15428 | accepted to EMNLP 2025 Findings | Data & Code