未验证 提交 cae6614c 编写于 作者: Z Zeng Jinle 提交者: GitHub

Merge pull request #16005 from sneaxiy/develop-fix-executor-doc

Fix develop Executor doc
...@@ -261,45 +261,42 @@ def _as_lodtensor(data, place): ...@@ -261,45 +261,42 @@ def _as_lodtensor(data, place):
class Executor(object): class Executor(object):
""" """
An Executor in Python, only support the single-GPU running. For multi-cards, please refer to An Executor in Python, supports single/multiple-GPU running, and single/multiple-CPU running.
ParallelExecutor. Python executor takes a program, adds feed operators and fetch operators to this program according
Python executor takes a program, add feed operators and fetch operators to this program according
to feed map and fetch_list. Feed map provides input data for the program. fetch_list provides to feed map and fetch_list. Feed map provides input data for the program. fetch_list provides
the variables(or names) that user want to get after program run. Note: the executor will run all the variables(or names) that user wants to get after program runs. Note: the executor will run all
operators in the program but not only the operators dependent by the fetch_list. operators in the program but not only the operators dependent by the fetch_list.
It store the global variables into the global scope, and create a local scope for the temporary It stores the global variables into the global scope, and creates a local scope for the temporary
variables. The local scope contents will be discarded after every minibatch forward/backward finished. variables. The contents in local scope may be discarded after every minibatch forward/backward
But the global scope variables will be persistent through different runs. finished. But the global scope variables will be persistent through different runs.
All of ops in program will be running in sequence.
Example: Example:
.. code-block:: python
# First create the Executor. .. code-block:: python
place = fluid.CUDAPlace(0) if use_cuda else fluid.CPUPlace()
exe = fluid.Executor(place) # First create the Executor.
place = fluid.CUDAPlace(0) if use_cuda else fluid.CPUPlace()
# Run the startup program once and only once. exe = fluid.Executor(place)
# Not need to optimize/compile the startup program.
exe.run(fluid.default_startup_program()) # Run the startup program once and only once.
# Not need to optimize/compile the startup program.
# Run the main program directly without compile. exe.run(fluid.default_startup_program())
loss, = exe.run(fluid.default_main_program(),
feed=feed_dict, # Run the main program directly without compile.
fetch_list=[loss.name]) loss, = exe.run(fluid.default_main_program(),
# Or, compiled the program and run. See `CompiledProgram` for more detail. feed=feed_dict,
compiled_prog = compiler.CompiledProgram( fetch_list=[loss.name])
fluid.default_main_program()).with_data_parallel( # Or, compiled the program and run. See `CompiledProgram` for more detail.
loss_name=loss.name) compiled_prog = compiler.CompiledProgram(
loss, = exe.run(compiled_prog, fluid.default_main_program()).with_data_parallel(
feed=feed_dict, loss_name=loss.name)
fetch_list=[loss.name]) loss, = exe.run(compiled_prog,
feed=feed_dict,
fetch_list=[loss.name])
Args: Args:
place(core.CPUPlace|core.CUDAPlace(n)): indicate the executor run on which device place(core.CPUPlace|core.CUDAPlace(n)): indicate the executor run on which device
Note: For debugging complicated network in parallel-GPUs, you can test it on the executor.
They has the exactly same arguments, and expected the same results.
""" """
def __init__(self, place): def __init__(self, place):
...@@ -382,6 +379,12 @@ class Executor(object): ...@@ -382,6 +379,12 @@ class Executor(object):
] ]
return outs return outs
'''
TODO(typhoonzero): Define "no longer use" meaning? Can user create
a new Executor for the same program and run?
TODO(panyx0718): Why ParallelExecutor doesn't have close?
'''
def close(self): def close(self):
""" """
Close this executor. Close this executor.
...@@ -389,9 +392,6 @@ class Executor(object): ...@@ -389,9 +392,6 @@ class Executor(object):
You can no longer use this executor after calling this method. You can no longer use this executor after calling this method.
For the distributed training, this method would free the resource on PServers related to For the distributed training, this method would free the resource on PServers related to
the current Trainer. the current Trainer.
TODO(typhoonzero): Define "no longer use" meaning? Can user create
a new Executor for the same program and run?
TODO(panyx0718): Why ParallelExecutor doesn't have close?
Example: Example:
>>> cpu = core.CPUPlace() >>> cpu = core.CPUPlace()
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册