remove redundant command argument for training using launch.py (PaddlePaddle#1049)

remove command argument train.py from the document, which is added by mistake.

Author: lilong12

doc/fluid/user_guides/howto/training/cluster_howto.rst

@@ -247,7 +247,7 @@ NCCL2模式必要参数说明
 
 .. code-block:: bash
 
-    > PADDLE_NUM_TRAINERS=<TRAINER_COUNT> python -m paddle.distributed.launch train.py --gpus <NUM_GPUS_ON_HOSTS> <ENTRYPOINT_SCRIPT> --arg1 --arg2 ...
+    > PADDLE_NUM_TRAINERS=<TRAINER_COUNT> python -m paddle.distributed.launch --gpus <NUM_GPUS_ON_HOSTS> <ENTRYPOINT_SCRIPT> --arg1 --arg2 ...
 
 
 NCCL2分布式训练注意事项

doc/fluid/user_guides/howto/training/cluster_howto_en.rst

@@ -226,7 +226,7 @@ Attention during usage:
 
   .. code-block:: bash
 
-     > PADDLE_NUM_TRAINERS=<TRAINER_COUNT> python -m paddle.distributed.launch train.py --gpus <NUM_GPUS_ON_HOSTS> <ENTRYPOINT_SCRIPT> --arg1 --arg2 ...
+     > PADDLE_NUM_TRAINERS=<TRAINER_COUNT> python -m paddle.distributed.launch --gpus <NUM_GPUS_ON_HOSTS> <ENTRYPOINT_SCRIPT> --arg1 --arg2 ...
 
 Important Notes on NCCL2 Distributed Training
 ++++++++++++++++++++++++++++++++++++++++++++++