When the Same Coefficients Reach Different Places: Asymmetric Realizability in Transplanting Tokenizers across Large Language Models 事件

When the Same Coefficients Reach Different Places: Asymmetric Realizability in Transplanting Tokenizers across Large Language Models arXiv:2601.00065v3 Announce Type: replace-cross Abstract: Tokenizer transplant in cross-vocabulary model composition reconstructs donor-only embedding rows as weighted combinations over shared lexical anchors and reuses those coefficients on the base. We identify a structural geometric property of this reconstruction: the same coefficient vector reaches different