Merge pull request #12 from PaddlePaddle/develop

Develop
PaddlePaddle · Apr 3, 2019 · 57f14fe · 57f14fe
2 parents 46e1140 + 54699e8
commit 57f14fe
Show file tree

Hide file tree

Showing 19 changed files with 22 additions and 19 deletions.
diff --git a/ce_models/mnist/__init__.py b/ce_models/mnist/__init__.py
@@ -0,0 +1 @@
+
diff --git a/ce_models/resnet50/latest_kpis/cifar10_128_gpu_memory_factor.txt b/ce_models/resnet50/latest_kpis/cifar10_128_gpu_memory_factor.txt
@@ -1 +1 @@
-[1394]
+[1342]
diff --git a/ce_models/resnet50/latest_kpis/cifar10_128_train_speed_factor.txt b/ce_models/resnet50/latest_kpis/cifar10_128_train_speed_factor.txt
@@ -1 +1 @@
-[767.2907104492188]
+[2310.2907104492188]
diff --git a/ce_models/resnet50/latest_kpis/flowers_64_gpu_memory_factor.txt b/ce_models/resnet50/latest_kpis/flowers_64_gpu_memory_factor.txt
@@ -1 +1 @@
-[10352]
+[8732]
diff --git a/ce_models/resnet50/latest_kpis/flowers_64_train_speed_factor.txt b/ce_models/resnet50/latest_kpis/flowers_64_train_speed_factor.txt
@@ -1 +1 @@
-[92.87747192382812]
+[193.87747192382812]
diff --git a/ce_models/resnet50/model.py b/ce_models/resnet50/model.py
@@ -16,6 +16,7 @@
 
 import paddle
 import paddle.fluid as fluid
+import paddle.fluid.compiler as compiler
 import paddle.fluid.core as core
 import paddle.fluid.profiler as profiler
 
@@ -213,12 +214,11 @@ def run_benchmark(model, args):
         input=predict, label=label, total=batch_size_tensor)
 
     inference_program = fluid.default_main_program().clone(for_test=True)
+    inference_program = compiler.CompiledProgram(inference_program).with_data_parallel()
 
     optimizer = fluid.optimizer.Momentum(learning_rate=0.01, momentum=0.9)
     opts = optimizer.minimize(avg_cost)
 
-    fluid.memory_optimize(fluid.default_main_program())
-
     train_reader = paddle.batch(
         paddle.reader.shuffle(
             paddle.dataset.cifar.train10()
@@ -249,6 +249,8 @@ def test(exe):
     place = core.CPUPlace() if args.device == 'CPU' else core.CUDAPlace(0)
     exe = fluid.Executor(place)
     exe.run(fluid.default_startup_program())
+    compiled_prog = compiler.CompiledProgram(fluid.default_main_program()).with_data_parallel(
+        loss_name=avg_cost.name)
     accuracy = fluid.average.WeightedAverage()
     if args.use_fake_data:
         data = train_reader().next()
@@ -285,7 +287,7 @@ def test(exe):
                 label = np.array(map(lambda x: x[1], data)).astype('int64')
                 label = label.reshape([-1, 1])
             loss, acc, weight = exe.run(
-                fluid.default_main_program(),
+                compiled_prog,
                 feed={'data': image,
                       'label': label},
                 fetch_list=[avg_cost, batch_acc, batch_size_tensor])

diff --git a/ce_models/resnet50_net_CPU/train.py b/ce_models/resnet50_net_CPU/train.py
@@ -178,7 +178,8 @@ def get_parallel_executor(args, avg_cost, train_program, test_program):
     build_strategy.reduce_strategy = fluid.BuildStrategy.ReduceStrategy.Reduce \
         if args.reduce_strategy == "Reduce" \
         else fluid.BuildStrategy.ReduceStrategy.AllReduce
-
+    if args.reduce_strategy == "Reduce":
+        build_strategy.fuse_broadcast_ops = True
     train_exe = fluid.ParallelExecutor(
         loss_name=avg_cost.name,
         main_program=train_program,

diff --git a/ce_models/vgg16/latest_kpis/cifar10_128_train_speed_factor.txt b/ce_models/vgg16/latest_kpis/cifar10_128_train_speed_factor.txt
@@ -1 +1 @@
-[730.0634155273438]
+[1487.0634155273438]
diff --git a/ce_models/vgg16/latest_kpis/flowers_32_gpu_memory_factor.txt b/ce_models/vgg16/latest_kpis/flowers_32_gpu_memory_factor.txt
@@ -1 +1 @@
-[10547]
+[8902]
diff --git a/ce_models/vgg16/latest_kpis/flowers_32_train_speed_factor.txt b/ce_models/vgg16/latest_kpis/flowers_32_train_speed_factor.txt
@@ -1 +1 @@
-[68.13056182861328]
+[105.13056182861328]
diff --git a/models/human_pose_estimation/latest_kpis/each_pass_duration_card1_factor.txt b/models/human_pose_estimation/latest_kpis/each_pass_duration_card1_factor.txt
@@ -1 +1 @@
-[4.80402100086]
+[23.6]
diff --git a/models/human_pose_estimation/latest_kpis/each_pass_duration_card4_factor.txt b/models/human_pose_estimation/latest_kpis/each_pass_duration_card4_factor.txt
@@ -1 +1 @@
-[3.33657252789]
+[20.1766821543]
diff --git a/models/human_pose_estimation/latest_kpis/each_pass_duration_card8_factor.txt b/models/human_pose_estimation/latest_kpis/each_pass_duration_card8_factor.txt
diff --git a/models/human_pose_estimation/latest_kpis/train_loss_card1_factor.txt b/models/human_pose_estimation/latest_kpis/train_loss_card1_factor.txt
@@ -1 +1 @@
-[0.0014293382]
+[0.00131]
diff --git a/models/human_pose_estimation/latest_kpis/train_loss_card4_factor.txt b/models/human_pose_estimation/latest_kpis/train_loss_card4_factor.txt
@@ -1 +1 @@
-[0.0012988691]
+[0.0012868968]
diff --git a/models/human_pose_estimation/latest_kpis/train_loss_card8_factor.txt b/models/human_pose_estimation/latest_kpis/train_loss_card8_factor.txt
diff --git a/models/image_classification/latest_kpis/test_acc_top1_factor.txt b/models/image_classification/latest_kpis/test_acc_top1_factor.txt
@@ -1 +1 @@
-[0.232039]
+[0.204901]
diff --git a/models/image_classification/latest_kpis/test_cost_factor.txt b/models/image_classification/latest_kpis/test_cost_factor.txt
@@ -1 +1 @@
-[3.1024718]
+[3.0848286]
diff --git a/test_models/__init__.py b/test_models/__init__.py
@@ -0,0 +1 @@
+