I used below code to train my model in v3-8 TPU which is single device TPU, it works fine, however, the same code does not work on TPU pod v2-32. As I know v2-32 are clusters of
