Tensorflow out of memory (OOM) with multiple tape.gradient calls

Question

I am running a custom training loop in tensorflow 2 that combines two different losses using two different calls to tape.gradient(), like so:

<