TVM trains the model like tensorflow1

The training of large models is generally more stable and faster using static images, which is also the strength of TVM. I would like to know if the community would consider allowing TVM to support training like tensorflow1.