We need a robust strategy for handling bad requests in each backend. For example, if a request comes with the wrong data type or shape, it needs to be handled and then not passed on to the responder. What should happen if a subset of requests in a batch are bad?
We need a robust strategy for handling bad requests in each backend. For example, if a request comes with the wrong data type or shape, it needs to be handled and then not passed on to the responder. What should happen if a subset of requests in a batch are bad?