follow comments

benchmark/fluid/fluid_benchmark.py

@@ -38,7 +38,10 @@ def parse_args():
         default='resnet',
         help='The model to run benchmark with.')
     parser.add_argument(
-        '--batch_size', type=int, default=32, help='The minibatch size.')
+        '--batch_size',
+        type=int,
+        default=32,
+        help='The batch size on each gpu.')
     parser.add_argument(
         '--learning_rate', type=float, default=0.001, help='The learning rate.')
     parser.add_argument(
@@ -229,27 +232,35 @@ def train(avg_loss, infer_prog, optimizer, train_reader, test_reader, batch_acc,
     iters, num_samples, start_time = 0, 0, time.time()
     for pass_id in range(args.pass_num):
         train_losses = []
+        if not args.use_reader_op:
             reader_generator = train_reader()
         batch_id = 0
         data = None
         while True:
             if not args.use_reader_op:
                 data = next(reader_generator, None)
-            if iters == args.iterations or data == None:
+                if data == None:
+                    break
+            if iters == args.iterations:
                 break
             if iters == args.skip_batch_num:
                 start_time = time.time()
                 num_samples = 0
 
             if args.use_reader_op:
+                try:
                     loss = exe.run(train_prog, fetch_list=[avg_loss])
+                except fluid.core.EnforceNotMet as ex:
+                    break
             else:
                 loss = exe.run(train_prog,
                                feed=feeder.feed(data),
                                fetch_list=[avg_loss])
             iters += 1
             batch_id += 1
-            # FIXME(wuyi): last batch size maybe different
+            # FIXME(wuyi): For use_reader_op, if the current
+            # pass is not the last, the last batch of this pass
+            # is also equal to args.batch_size.
             num_samples += len(args.batch_size)
             train_losses.append(loss)
             print("Pass: %d, Iter: %d, Loss: %f\n" %
@@ -315,13 +326,16 @@ def train_parallel(avg_loss, infer_prog, optimizer, train_reader, test_reader,
         num_samples = 0
         iters = 0
         start_time = time.time()
+        if not args.use_reader_op:
             reader_generator = train_reader()
         batch_id = 0
         data = None
         while True:
             if not args.use_reader_op:
                 data = next(reader_generator, None)
-            if iters == args.iterations or data == None:
+                if data == None:
+                    break
+            if iters == args.iterations:
                 break
             if args.profile and pass_id == 0 and batch_id == 5:
                 profiler.start_profiler("All")
@@ -335,7 +349,10 @@ def train_parallel(avg_loss, infer_prog, optimizer, train_reader, test_reader,
             if args.use_reader_op and iters >= args.iterations / args.gpus:
                 break
             if args.use_fake_data or args.use_reader_op:
+                try:
                     loss, = exe.run([avg_loss.name])
+                except fluid.core.EnforceNotMet as ex:
+                    break
             else:
                 loss, = exe.run([avg_loss.name], feed=feeder.feed(data))
             if args.update_method == "pserver":

benchmark/fluid/models/machine_translation.py

@@ -223,7 +223,7 @@ def get_model(args):
     train_batch_generator = paddle.batch(
         paddle.reader.shuffle(
             paddle.dataset.wmt14.train(dict_size), buf_size=1000),
-        batch_size=args.batch_size)
+        batch_size=args.batch_size * args.gpus)
 
     test_batch_generator = paddle.batch(
         paddle.reader.shuffle(

benchmark/fluid/models/mnist.py

@@ -103,7 +103,7 @@ def get_model(args):
 
     # Reader
     train_reader = paddle.batch(
-        paddle.dataset.mnist.train(), batch_size=args.batch_size)
+        paddle.dataset.mnist.train(), batch_size=args.batch_size * args.gpus)
     test_reader = paddle.batch(
         paddle.dataset.mnist.test(), batch_size=args.batch_size)
     return avg_cost, inference_program, opt, train_reader, test_reader, batch_acc

benchmark/fluid/models/resnet.py

@@ -184,7 +184,7 @@ def get_model(args):
     batched_train_reader = paddle.batch(
         paddle.reader.shuffle(
             train_reader, buf_size=5120),
-        batch_size=args.batch_size)
+        batch_size=args.batch_size * args.gpus)
     batched_test_reader = paddle.batch(train_reader, batch_size=args.batch_size)
 
     return avg_cost, inference_program, optimizer, batched_train_reader, batched_test_reader, batch_acc

benchmark/fluid/models/stacked_dynamic_lstm.py

@@ -118,7 +118,7 @@ def get_model(args):
     train_reader = batch(
         paddle.reader.shuffle(
             crop_sentence(imdb.train(word_dict), crop_size), buf_size=25000),
-        batch_size=args.batch_size)
+        batch_size=args.batch_size * args.gpus)
     test_reader = batch(
         paddle.reader.shuffle(
             crop_sentence(imdb.test(word_dict), crop_size), buf_size=25000),

benchmark/fluid/models/vgg.py

@@ -110,7 +110,7 @@ def get_model(args):
             paddle.dataset.cifar.train10()
             if args.data_set == 'cifar10' else paddle.dataset.flowers.train(),
             buf_size=5120),
-        batch_size=args.batch_size)
+        batch_size=args.batch_size * args.gpus)
     test_reader = paddle.batch(
         paddle.dataset.cifar.test10()
         if args.data_set == 'cifar10' else paddle.dataset.flowers.test(),