Transporting Task Vectors across Different Architectures without Training

ArXi:2602.12952v2 Announce Type: replace-cross Adapting large pre-trained models to downstream tasks often produces task-specific parameter updates that are expensive to relearn for every model variant. While recent work has shown that such updates can be transferred between models with identical architectures, transferring them across models of different widths remains unexplored. In this work, we