docker model package doesn't inherent model features ( no vision decoding )

[Glog78]

original model from dockerhub:

docker model rm gemma4:12B_256k
docker model package --from ai/gemma4:12B --context-size 262144 gemma4:12B_256k
docker model run gemma4:12B_256k ~/Bilder/Screenshots/axu_hochkant.png whats on the screenshot ?
docker model run gemma4:12B_256k who are you ?on the screenshot ?

^^ as expected the original model starts to describe the image