Batch size TPUv4

I am running the run_mlm_flax-scripts on TPUv4. Will the parameter per_device_batch_size automatically scale when running on pods, or do the patitions needs to be defined? And how? To clarify: Lets say I am running a script with per_devic_batch_size=100 on 4 TPUs (a TPUv4-32). How big will the actual batch size be?