# TODO: remove this decorator after we finalize training API
def __impl__(*args, **kwargs):
program_translator = ProgramTranslator()
if in_dygraph_mode() or not program_translator.enable_declarative:
logger.info(
"The decorator 'dygraph_to_static_func' doesn't work in "
"dygraph mode or set ProgramTranslator.enable to False. "
"We will just return dygraph output.")
return dygraph_func(*args, **kwargs)
static_func = program_translator.get_func(dygraph_func)
return static_func(*args, **kwargs)
return __impl__
dygraph_to_static_func = wrap_decorator(_dygraph_to_static_func_)
def _declarative_(dygraph_func):
"""
Converts imperative dygraph APIs into declarative function APIs. Decorator
@declarative handles the Program and Executor of static mode and returns
the result as a dygraph VarBase.
Args:
dygraph_func (callable): callable imperative function.
Returns:
VarBase: containing the numerical result.
Examples:
__all__ = ["DistributedStrategy"]
non_auto_func_called = True
def __non_auto_func_called__(func):
def __impl__(*args, **kwargs):
global non_auto_func_called
non_auto_func_called = False
return func(*args, **kwargs)
return __impl__
is_strict_auto = wrap_decorator(__non_auto_func_called__)
def get_msg_dict(msg):
res_dict = {}
fields = msg.DESCRIPTOR.fields
for f in fields:
res_dict[f.name] = getattr(msg, f.name)
return res_dict
def assign_configs_value(msg, config):
fields = msg.DESCRIPTOR.fields
for key in config:
for f in fields:
if key == f.name:
import numpy as np
import paddle
def _dygraph_guard_(func):
def __impl__(*args, **kwargs):
if paddle.in_dynamic_mode():
return func(*args, **kwargs)
else:
with fluid.dygraph.guard():
return func(*args, **kwargs)
return __impl__
dygraph_guard = wrap_decorator(_dygraph_guard_)
def random_var(size, low=-1, high=1, dtype='float32'):
x_np = np.random.uniform(low=low, high=high, size=size).astype(dtype)
return fluid.dygraph.to_variable(x_np)
class TestDygraphDoubleGrad(TestCase):
def setUp(self):
self.sort_sum_gradient = False
self.shape = [5, 10]
def grad(self,
outputs,
inputs,
def __impl__(*args, **kwargs):
cls = args[0]
if cls._role_maker is not None and cls._role_maker._is_non_distributed(
) is True:
warnings.warn(
"%s() function doesn't work when use non_distributed fleet." %
(func.__name__))
return
return func(*args, **kwargs)
return __impl__
inited_runtime_handler = wrap_decorator(_inited_runtime_handler_)
is_non_distributed_check = wrap_decorator(_is_non_distributed_check_)
class Fleet(object):
"""
Unified API for distributed training of PaddlePaddle
Please reference the https://github.com/PaddlePaddle/FleetX for details
Returns:
Fleet: A Fleet instance
Example for collective training:
.. code-block:: python