AI RESEARCH
When the Same Coefficients Reach Different Places: Asymmetric Realizability in Transplanting Tokenizers across Large Language Models
arXiv CS.LG
•
ArXi:2601.00065v3 Announce Type: replace 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 sets in the donor and base anchor spans, an \emph{asymmetric realizability} gap.