Computes updates on 'vmap' to optimize 'loss'. TODO(rpang): explore merging gradient_mask and gradient_adjuster. Args: metrics: A Dict[str, (value, weight)], from which loss can be extracted according to p.loss_name. vmap: A `.NestedMap` object containing variables to o
(self, metrics, vmap, gradient_mask=None, gradient_adjuster=None)
source not stored for this graph (policy: none)
nothing calls this directly
no test coverage detected