Hi,
Sorry for the late. Based on the source, the summary do add in the eager training mode:
Have you checked this issue with TensorFlow team? They may know more about the difference in the summary API.
Thanks.