update by comment

fluid/image_classification/dist_train/README.md

 # Distributed Image Classification Models Training
 
 This folder contains implementations of **Image Classification Models**, they are designed to support
-large-scaled distributed training with two distributed architecture: parameter server with RPC and ring-base with Nvidia NCCL2 library.
+large-scaled distributed training with two distributed mode: parameter server mode and NCCL2(Nvidia NCCL2 communication library) collective mode.
 
 ## Getting Started
 
-Before getting started, please make sure you have finished the imagenet [Data Preparation](../README.md#data-preparation).
+Before getting started, please make sure you have go throught the imagenet [Data Preparation](../README.md#data-preparation).
 
 1. The entrypoint file is `dist_train.py`, some important flags are as follows:
 
     - `model`, the model to run with, such as `ResNet50`, `ResNet101` and etc..
     - `batch_size`, the batch_size per device.
-    - `update_method`, specify the update method, local, pserver or nccl2.
+    - `update_method`, specify the update method, can choose from local, pserver or nccl2.
     - `device`, use CPU or GPU device.
     - `gpus`, the GPU device count that the process used.
 
@@ -29,7 +29,7 @@ Before getting started, please make sure you have finished the imagenet [Data Pr
     - `PADDLE_PSERVER_PORT`, the port of the parameter pserver listened on.
     - `PADDLE_TRAINER_IPS`, the trainer IP list, separated by ",", only be used with upadte_method is nccl2.
 
-### Pserver Server with RPC
+### Parameter Server Mode
 
 In this example, we launched 4 parameter server instances and 4 trainer instances in the cluster:
 
@@ -66,7 +66,7 @@ In this example, we launched 4 parameter server instances and 4 trainer instance
 
     ```
 
-### Ring-base with Nvidia NCCL2 library
+### NCCL2 Collective Mode
 
 1. launch trainer process
 
@@ -83,9 +83,9 @@ In this example, we launched 4 parameter server instances and 4 trainer instance
         --data_dir=../data/ILSVRC2012
     ```
 
-### Training Curve
+### Visualize the Training Process
 
-It's easy to draw the training curve accroding to the training logs, for example,
+It's easy to draw the learning curve accroding to the training logs, for example,
 the logs of ResNet50 is as follows:
 
 ``` text

fluid/image_classification/dist_train/args.py

@@ -22,7 +22,7 @@ BENCHMARK_MODELS = [
 
 
 def parse_args():
-    parser = argparse.ArgumentParser('Fluid model benchmarks.')
+    parser = argparse.ArgumentParser('Distributed Image Classification Training.')
     parser.add_argument(
         '--model',
         type=str,
@@ -74,8 +74,6 @@ def parse_args():
         default='flowers',
         choices=['cifar10', 'flowers', 'imagenet'],
         help='Optional dataset for benchmark.')
-    parser.add_argument(
-        '--infer_only', action='store_true', help='If set, run forward only.')
     parser.add_argument(
         '--no_test',
         action='store_true',
@@ -84,10 +82,6 @@ def parse_args():
         '--memory_optimize',
         action='store_true',
         help='If set, optimize runtime memory before start.')
-    parser.add_argument(
-        '--use_fake_data',
-        action='store_true',
-        help='If set ommit the actual read data operators.')
     parser.add_argument(
         '--update_method',
         type=str,
@@ -104,19 +98,10 @@ def parse_args():
         action='store_true',
         default=False,
         help='Whether start pserver in async mode to support ASGD')
-    parser.add_argument(
-        '--use_reader_op',
-        action='store_true',
-        help='Whether to use reader op, and must specify the data path if set this to true.'
-    )
     parser.add_argument(
         '--no_random',
         action='store_true',
         help='If set, keep the random seed and do not shuffle the data.')
-    parser.add_argument(
-        '--use_lars',
-        action='store_true',
-        help='If set, use lars for optimizers, ONLY support resnet module.')
     parser.add_argument(
         '--reduce_strategy',
         type=str,