fix infer_from_dataset docs and abastract class problem

test=develop

python/paddle/fluid/device_worker.py

@@ -17,7 +17,7 @@ __all__ = ['DeviceWorker', 'Hogwild', 'DownpourSGD']
 
 class DeviceWorker(object):
     """
-    DeviceWorker is a abstract class, which generates worker desc.
+    DeviceWorker is an abstract class, which generates worker desc.
     This class is an inner class that we do computation logics within
     the implementation. For example, execution of a program or a graph.
     """

python/paddle/fluid/executor.py

@@ -671,6 +671,7 @@ class Executor(object):
         push gradients will be disabled in infer_from_dataset.
         infer_from_dataset() can be used for evaluation in multi-thread
         very easily.
+
         Args:
             program(Program|CompiledProgram): the program that needs to be run,
                if not provided, then default_main_program (not compiled) will be used.