Patching open-vocabulary models by interpolating weights

Open in new window