1 paper · 1 filter
Szczepan Konior, Alexandre Quemy, PrzemysÅaw Klocek +2
Understanding the geometric structure of pre-trained language model embeddings matters for interpretability and safety. We ask whether sentence-level classification signal lives in…