Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into format

Author: weixing02

benchmark/.gitignore

@@ -7,3 +7,6 @@ paddle/rnn/imdb.pkl
 caffe/image/logs
 tensorflow/image/logs
 tensorflow/rnn/logs
+fluid/models/*.pyc
+fluid/logs
+fluid/nohup.out

benchmark/fluid/fluid_benchmark.py

@@ -40,10 +40,7 @@ def parse_args():
     parser.add_argument(
         '--batch_size', type=int, default=32, help='The minibatch size.')
     parser.add_argument(
-        '--learning_rate',
-        type=float,
-        default=0.001,
-        help='The minibatch size.')
+        '--learning_rate', type=float, default=0.001, help='The learning rate.')
     # TODO(wuyi): add "--use_fake_data" option back.
     parser.add_argument(
         '--skip_batch_num',
@@ -88,8 +85,8 @@ def parse_args():
         help='If set, use nvprof for CUDA.')
     parser.add_argument(
         '--no_test',
-        action='store_false',
-        help='If set, test the testset during training.')
+        action='store_true',
+        help='If set, do not test the testset during training.')
     parser.add_argument(
         '--memory_optimize',
         action='store_true',
@@ -231,13 +228,10 @@ def train(avg_loss, infer_prog, optimizer, train_reader, test_reader, batch_acc,
             train_losses.append(loss)
             print("Pass: %d, Iter: %d, Loss: %f\n" %
                   (pass_id, iters, np.mean(train_losses)))
-        train_elapsed = time.time() - start_time
-        examples_per_sec = num_samples / train_elapsed
-        print('\nTotal examples: %d, total time: %.5f, %.5f examples/sec\n' %
-              (num_samples, train_elapsed, examples_per_sec))
-        print("Pass: %d, Loss: %f" % (pass_id, np.mean(train_losses)))
+        print_train_time(start_time, time.time(), num_samples)
+        print("Pass: %d, Loss: %f" % (pass_id, np.mean(train_losses))),
         # evaluation
-        if not args.no_test and batch_acc != None:
+        if not args.no_test and batch_acc:
             pass_test_acc = test(exe, infer_prog, test_reader, feeder,
                                  batch_acc)
             print(", Test Accuracy: %f" % pass_test_acc)
@@ -315,11 +309,8 @@ def train_parallel(avg_loss, infer_prog, optimizer, train_reader, test_reader,
             if batch_id % 1 == 0:
                 print("Pass %d, batch %d, loss %s" %
                       (pass_id, batch_id, np.array(loss)))
-        train_elapsed = time.time() - start_time
-        examples_per_sec = num_samples / train_elapsed
-        print('\nTotal examples: %d, total time: %.5f, %.5f examples/sed\n' %
-              (num_samples, train_elapsed, examples_per_sec))
-        if not args.no_test and batch_acc != None:
+        print_train_time(start_time, time.time(), num_samples)
+        if not args.no_test and batch_acc:
             test_acc = test(startup_exe, infer_prog, test_reader, feeder,
                             batch_acc)
             print("Pass: %d, Test Accuracy: %f\n" % (pass_id, test_acc))
@@ -329,20 +320,27 @@ def train_parallel(avg_loss, infer_prog, optimizer, train_reader, test_reader,
 def print_arguments(args):
     vars(args)['use_nvprof'] = (vars(args)['use_nvprof'] and
                                 vars(args)['device'] == 'GPU')
-    print('----------- resnet Configuration Arguments -----------')
+    print('----------- Configuration Arguments -----------')
     for arg, value in sorted(vars(args).iteritems()):
         print('%s: %s' % (arg, value))
     print('------------------------------------------------')
 
 
+def print_train_time(start_time, end_time, num_samples):
+    train_elapsed = end_time - start_time
+    examples_per_sec = num_samples / train_elapsed
+    print('\nTotal examples: %d, total time: %.5f, %.5f examples/sed\n' %
+          (num_samples, train_elapsed, examples_per_sec))
+
+
 def main():
     args = parse_args()
     print_arguments(args)
 
     # the unique trainer id, starting from 0, needed by trainer
     # only
     nccl_id_var, num_trainers, trainer_id = (
-        None, 1, int(os.getenv("PADDLE_TRAINER_ID", "-1")))
+        None, 1, int(os.getenv("PADDLE_TRAINER_ID", "0")))
 
     if args.use_cprof:
         pr = cProfile.Profile()

benchmark/fluid/run.sh

@@ -2,6 +2,7 @@
 # This script benchmarking the PaddlePaddle Fluid on
 # single thread single GPU.
 
+mkdir -p logs
 #export FLAGS_fraction_of_gpu_memory_to_use=0.0
 export CUDNN_PATH=/paddle/cudnn_v5
 
@@ -35,6 +36,7 @@ nohup stdbuf -oL nvidia-smi \
       --format=csv \
       --filename=mem.log  \
       -l 1 &
+
 # mnist
 # mnist gpu mnist 128
 FLAGS_benchmark=true stdbuf -oL python fluid_benchmark.py \
@@ -43,7 +45,7 @@ FLAGS_benchmark=true stdbuf -oL python fluid_benchmark.py \
                --batch_size=128 \
                --skip_batch_num=5 \
                --iterations=500 \
-               2>&1 | tee -a mnist_gpu_128.log
+               2>&1 | tee -a logs/mnist_gpu_128.log
 
 # vgg16
 # gpu cifar10 128
@@ -53,7 +55,7 @@ FLAGS_benchmark=true stdbuf -oL python fluid_benchmark.py \
                --batch_size=128 \
                --skip_batch_num=5 \
                --iterations=30 \
-               2>&1 | tee -a vgg16_gpu_128.log
+               2>&1 | tee -a logs/vgg16_gpu_128.log
 
 # flowers gpu  128
 FLAGS_benchmark=true stdbuf -oL python fluid_benchmark.py \
@@ -63,28 +65,28 @@ FLAGS_benchmark=true stdbuf -oL python fluid_benchmark.py \
                --data_set=flowers \
                --skip_batch_num=5 \
                --iterations=30 \
-               2>&1 | tee -a vgg16_gpu_flowers_32.log
+               2>&1 | tee -a logs/vgg16_gpu_flowers_32.log
 
 # resnet50
 # resnet50 gpu cifar10 128
 FLAGS_benchmark=true stdbuf -oL python fluid_benchmark.py \
-               --model=resnet50 \
+               --model=resnet \
                --device=GPU \
                --batch_size=128 \
                --data_set=cifar10 \
                --skip_batch_num=5 \
                --iterations=30 \
-               2>&1 | tee -a resnet50_gpu_128.log
+               2>&1 | tee -a logs/resnet50_gpu_128.log
 
 # resnet50 gpu flowers 64
 FLAGS_benchmark=true stdbuf -oL python fluid_benchmark.py \
-               --model=resnet50 \
+               --model=resnet \
                --device=GPU \
                --batch_size=64 \
                --data_set=flowers \
                --skip_batch_num=5 \
                --iterations=30 \
-               2>&1 | tee -a resnet50_gpu_flowers_64.log
+               2>&1 | tee -a logs/resnet50_gpu_flowers_64.log
 
 # lstm
 # lstm gpu imdb 32 # tensorflow only support batch=32
@@ -94,7 +96,7 @@ FLAGS_benchmark=true stdbuf -oL python fluid_benchmark.py \
                --batch_size=32 \
                --skip_batch_num=5 \
                --iterations=30 \
-               2>&1 | tee -a lstm_gpu_32.log
+               2>&1 | tee -a logs/lstm_gpu_32.log
 
 # seq2seq
 # seq2seq gpu wmb 128
@@ -104,4 +106,4 @@ FLAGS_benchmark=true stdbuf -oL python fluid_benchmark.py \
                --batch_size=128 \
                --skip_batch_num=5 \
                --iterations=30 \
-               2>&1 | tee -a lstm_gpu_128.log
+               2>&1 | tee -a logs/lstm_gpu_128.log

cmake/external/grpc.cmake

@@ -33,10 +33,19 @@ ELSE()
   SET(BUILD_CMD make HAS_SYSTEM_PROTOBUF=false -s -j ${NUM_OF_PROCESSOR} static grpc_cpp_plugin)
 ENDIF()
 
+# FIXME(wuyi): do not build zlib cares protobuf twice, find a way to build grpc with them
 ExternalProject_Add(
     extern_grpc
     DEPENDS protobuf zlib
-    URL "http://paddlepaddledeps.bj.bcebos.com/grpc.tar.xz"
+    # NOTE(wuyi):
+    # this package is generated by following steps:
+    # 1. git clone -b v1.8.x https://github.com/grpc/grpc.git
+    # 2. submodule update --init
+    # 3. keep only zlib, cares, protobuf, boringssl under "third_party",
+    #    checkout and clean other dirs under third_party
+    # 4. remove .git, and package the directory.
+    URL "http://paddlepaddledeps.bj.bcebos.com/grpc-v1.8.x.tar.gz"
+    URL_MD5  "c9c58ee7d0e8929a63155af6a2ecdbd0"
     PREFIX          ${GRPC_SOURCES_DIR}
     UPDATE_COMMAND  ""
     CONFIGURE_COMMAND ""
@@ -49,7 +58,6 @@ ExternalProject_Add(
     INSTALL_COMMAND make prefix=${GRPC_INSTALL_DIR} install
 )
 
-# FIXME(typhoonzero): hack to get static lib path, try a better way like merge them.
 ADD_LIBRARY(grpc++_unsecure STATIC IMPORTED GLOBAL)
 SET_PROPERTY(TARGET grpc++_unsecure PROPERTY IMPORTED_LOCATION
              "${GRPC_INSTALL_DIR}/lib/libgrpc++_unsecure.a")

doc/fluid/api/io.rst

@@ -59,3 +59,21 @@ get_inference_program
 ..  autofunction:: paddle.fluid.io.get_inference_program
     :noindex:
 
+save_checkpoint
+---------------
+
+..  autofunction:: paddle.fluid.io.save_checkpoint
+    :noindex:
+
+load_checkpoint
+---------------
+
+..  autofunction:: paddle.fluid.io.load_checkpoint
+    :noindex:
+
+clean_checkpoint
+----------------
+
+..  autofunction:: paddle.fluid.io.clean_checkpoint
+    :noindex:
+

doc/fluid/api/layers.rst

@@ -181,6 +181,12 @@ Print
 ..  autofunction:: paddle.fluid.layers.Print
     :noindex:
 
+is_empty
+--------
+
+..  autofunction:: paddle.fluid.layers.is_empty
+    :noindex:
+
 device
 ======
 
@@ -255,6 +261,19 @@ double_buffer
 ..  autofunction:: paddle.fluid.layers.double_buffer
     :noindex:
 
+random_data_generator
+---------------------
+
+..  autofunction:: paddle.fluid.layers.random_data_generator
+    :noindex:
+
+Preprocessor
+------------
+
+..  autoclass:: paddle.fluid.layers.Preprocessor
+    :members:
+    :noindex:
+
 nn
 ==
 
@@ -594,6 +613,29 @@ roi_pool
 ..  autofunction:: paddle.fluid.layers.roi_pool
     :noindex:
 
+dice_loss
+---------
+
+..  autofunction:: paddle.fluid.layers.dice_loss
+    :noindex:
+
+resize_bilinear
+---------------
+
+..  autofunction:: paddle.fluid.layers.resize_bilinear
+    :noindex:
+
+gather
+------
+
+..  autofunction:: paddle.fluid.layers.gather
+    :noindex:
+
+random_crop
+-----------
+
+..  autofunction:: paddle.fluid.layers.random_crop
+    :noindex:
 
 ops
 ===
@@ -742,6 +784,12 @@ sum
 ..  autofunction:: paddle.fluid.layers.sum
     :noindex:
 
+shape
+-----
+
+..  autofunction:: paddle.fluid.layers.shape
+    :noindex:
+
 sigmoid
 -------
 
@@ -991,27 +1039,3 @@ zeros
 ..  autofunction:: paddle.fluid.layers.zeros
     :noindex:
 
-topk
-----
-
-..  autofunction:: paddle.fluid.layers.topk
-    :noindex:
-
-dice_loss
-----
-
-..  autofunction:: paddle.fluid.layers.dice_loss
-    :noindex:
-
-resize_bilinear
-____
-
-..  autofunction:: paddle.fluid.layers.resize_bilinear
-    :noindex:
-
-gather
-____
-
-..  autofunction:: paddle.fluid.layers.gather
-    :noindex:
-

doc/fluid/api/optimizer.rst

@@ -47,28 +47,6 @@ DecayedAdagrad
     :members:
     :noindex:
 
-Adadelta
------------------
-
-..  autoclass:: paddle.fluid.optimizer.Adadelta
-    :members:
-    :noindex:
-
-RMSProp
------------------
-
-..  autoclass:: paddle.fluid.optimizer.RMSProp
-    :members:
-    :noindex:
-
-ModelAverage
------------------
-
-..  autoclass:: paddle.fluid.optimizer.ModelAverage
-    :members:
-    :noindex:
-
-
 SGDOptimizer
 ------------
 
@@ -111,25 +89,31 @@ DecayedAdagradOptimizer
     :members:
     :noindex:
 
+RMSPropOptimizer
+----------------
 
-AdadeltaOptimizer
------------------
-
-..  autoclass:: paddle.fluid.optimizer.AdadeltaOptimizer
+..  autoclass:: paddle.fluid.optimizer.RMSPropOptimizer
     :members:
     :noindex:
 
+Adadelta
+--------
 
-RMSPropOptimizer
------------------
+..  autoclass:: paddle.fluid.optimizer.Adadelta
+    :members:
+    :noindex:
 
-..  autoclass:: paddle.fluid.optimizer.RMSPropOptimizer
+ModelAverage
+------------
+
+..  autoclass:: paddle.fluid.optimizer.ModelAverage
     :members:
     :noindex:
-    
+
 Optimizer
 ---------
 
 ..  autoclass:: paddle.fluid.optimizer.Optimizer
     :members:
     :noindex:
+