update train.py

dygraph/core/train.py

@@ -57,7 +57,7 @@ def train(model,
 
     if nranks > 1:
         strategy = fluid.dygraph.prepare_context()
-        model_parallel = fluid.dygraph.DataParallel(model, strategy)
+        ddp_model = fluid.dygraph.DataParallel(model, strategy)
 
     batch_sampler = DistributedBatchSampler(
         train_dataset, batch_size=batch_size, shuffle=True, drop_last=True)
@@ -86,10 +86,10 @@ def train(model,
             images = data[0]
             labels = data[1].astype('int64')
             if nranks > 1:
-                loss = model_parallel(images, labels)
-                loss = model_parallel.scale_loss(loss)
+                loss = ddp_model(images, labels)
+                loss = ddp_model.scale_loss(loss)
                 loss.backward()
-                model_parallel.apply_collective_grads()
+                ddp_model.apply_collective_grads()
             else:
                 loss = model(images, labels)
                 loss.backward()