A single train step. Apply back-propation and return the loss
(model, optimizer, train_iterator, args)
source not stored for this graph (policy: none)
no outgoing calls
no test coverage detected