32579822389423c7f4120e222aa26652f8507735,onmt/utils/optimizers.py,Optimizer,set_parameters,#Optimizer#Any#,172

Before Change

        self.sparse_params = []
        for k, p in params:
            if p.requires_grad:
                if self.method != "sparseadam" or "embed" not in k:
                    self.params.append(p)
                else:
                    self.sparse_params.append(p)

After Change

                if not param.requires_grad:
                    continue
                // TODO: Find a better way to check for sparse gradients.
                if "embed" in name:
                    sparse.append(param)
                else:
                    dense.append(param)
            self.optimizer = MultipleOptimizer(
                [optim.Adam(dense, lr=self.learning_rate,
                            betas=self.betas, eps=1e-8),
                 optim.SparseAdam(sparse, lr=self.learning_rate,

Instances