Skip to content

P
Paddle

BaiXuePrincess / Paddle
与 Fork 源项目一致

Fork自 PaddlePaddle / Paddle

通知 1
Star 1
Fork 0
P
Paddle
提交 ce468817 编写于 作者: Y Youwei Song 提交者： Aurelius84
浏览文件
操作

cherry-pick #20199 #20200 #20240 #20243 (#20410) 

test=release/1.6, test=document_fix
上级 3c30e92b
隐藏空白更改
内联 并排
Showing with 105 addition and 61 deletion
paddle/fluid/API.spec
浏览文件 @ ce468817
......@@ -11,10 +11,10 @@ paddle.fluid.default_startup_program (ArgSpec(args=[], varargs=None, keywords=No
paddle.fluid.default_main_program (ArgSpec(args=[], varargs=None, keywords=None, defaults=None), ('document', '853718df675e59aea7104f3d61bbf11d'))
paddle.fluid.program_guard (ArgSpec(args=['main_program', 'startup_program'], varargs=None, keywords=None, defaults=(None,)), ('document', '78fb5c7f70ef76bcf4a1862c3f6b8191'))
paddle.fluid.name_scope (ArgSpec(args=['prefix'], varargs=None, keywords=None, defaults=(None,)), ('document', '917d313881ff990de5fb18d98a9c7b42'))
paddle.fluid.cuda_places (ArgSpec(args=['device_ids'], varargs=None, keywords=None, defaults=(None,)), ('document', '1f2bb6ece651e44117652d2d7bedecf5'))
paddle.fluid.cpu_places (ArgSpec(args=['device_count'], varargs=None, keywords=None, defaults=(None,)), ('document', '956bab564ebc69ffd17195c08cc8ffa0'))
paddle.fluid.cuda_pinned_places (ArgSpec(args=['device_count'], varargs=None, keywords=None, defaults=(None,)), ('document', 'c2562241744aabe3fff1b59af22dd281'))
paddle.fluid.in_dygraph_mode (ArgSpec(args=[], varargs=None, keywords=None, defaults=None), ('document', '301bae0d8e02cc9eec5be02f052f11c6'))
paddle.fluid.cuda_places (ArgSpec(args=['device_ids'], varargs=None, keywords=None, defaults=(None,)), ('document', 'ab9bd2079536114aa7c1488a489ee87f'))
paddle.fluid.cpu_places (ArgSpec(args=['device_count'], varargs=None, keywords=None, defaults=(None,)), ('document', 'a7352a3dd39308fde4fbbf6421a4193d'))
paddle.fluid.cuda_pinned_places (ArgSpec(args=['device_count'], varargs=None, keywords=None, defaults=(None,)), ('document', '567ac29567716fd8e7432b533337d529'))
paddle.fluid.in_dygraph_mode (ArgSpec(args=[], varargs=None, keywords=None, defaults=None), ('document', 'df1f4d1ed7e1eefe04f6361efda6b75a'))
paddle.fluid.is_compiled_with_cuda (ArgSpec(args=[], varargs=None, keywords=None, defaults=None), ('document', '60c7f107a5050aeb58bb74eb175672b5'))
paddle.fluid.Variable ('paddle.fluid.framework.Variable', ('document', '65ff735c2b96673d7131f5ff6b0db40c'))
paddle.fluid.Variable.__init__ (ArgSpec(args=['self', 'block', 'type', 'name', 'shape', 'dtype', 'lod_level', 'capacity', 'persistable', 'error_clip', 'stop_gradient', 'is_data', 'need_check_feed'], varargs=None, keywords='kwargs', defaults=(VarType.LOD_TENSOR, None, None, None, None, None, None, None, False, False, False)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
......@@ -34,7 +34,7 @@ paddle.fluid.Executor.infer_from_dataset (ArgSpec(args=['self', 'program', 'data
paddle.fluid.Executor.run (ArgSpec(args=['self', 'program', 'feed', 'fetch_list', 'feed_var_name', 'fetch_var_name', 'scope', 'return_numpy', 'use_program_cache'], varargs=None, keywords=None, defaults=(None, None, None, 'feed', 'fetch', None, True, False)), ('document', '4cfcd9c15b766a51b584cc46d38f1ad8'))
paddle.fluid.Executor.train_from_dataset (ArgSpec(args=['self', 'program', 'dataset', 'scope', 'thread', 'debug', 'fetch_list', 'fetch_info', 'print_period', 'fetch_handler'], varargs=None, keywords=None, defaults=(None, None, None, 0, False, None, None, 100, None)), ('document', '73024c79f46b4f14f1060edeaa4919c8'))
paddle.fluid.global_scope (ArgSpec(args=[], varargs=None, keywords=None, defaults=None), ('document', 'f65788d9ead293ada47551339df12203'))
paddle.fluid.scope_guard (ArgSpec(args=['scope'], varargs=None, keywords=None, defaults=None), ('document', 'e6c073ed237001aaba7bff976b62b122'))
paddle.fluid.scope_guard (ArgSpec(args=['scope'], varargs=None, keywords=None, defaults=None), ('document', '02fcfc1eda07c03a84ed62422366239c'))
paddle.fluid.DistributeTranspiler ('paddle.fluid.transpiler.distribute_transpiler.DistributeTranspiler', ('document', 'b2b19821c5dffcd11473d6a4eef089af'))
paddle.fluid.DistributeTranspiler.__init__ (ArgSpec(args=['self', 'config'], varargs=None, keywords=None, defaults=(None,)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
paddle.fluid.DistributeTranspiler.get_pserver_program (ArgSpec(args=['self', 'endpoint'], varargs=None, keywords=None, defaults=None), ('document', 'b1951949c6d21698290aa8ac69afee32'))
......@@ -1041,7 +1041,7 @@ paddle.fluid.optimizer.RecomputeOptimizer.backward (ArgSpec(args=['self', 'loss'
paddle.fluid.optimizer.RecomputeOptimizer.get_opti_var_name_list (ArgSpec(args=['self'], varargs=None, keywords=None, defaults=None), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
paddle.fluid.optimizer.RecomputeOptimizer.load (ArgSpec(args=['self', 'stat_dict'], varargs=None, keywords=None, defaults=None), ('document', '7b2b8ae72011bc4decb67e97623f2c56'))
paddle.fluid.optimizer.RecomputeOptimizer.minimize (ArgSpec(args=['self', 'loss', 'startup_program', 'parameter_list', 'no_grad_set', 'grad_clip'], varargs=None, keywords=None, defaults=(None, None, None, None)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
paddle.fluid.backward.append_backward (ArgSpec(args=['loss', 'parameter_list', 'no_grad_set', 'callbacks', 'checkpoints'], varargs=None, keywords=None, defaults=(None, None, None, None)), ('document', '52488008103886c793843a3828bacd5e'))
paddle.fluid.backward.append_backward (ArgSpec(args=['loss', 'parameter_list', 'no_grad_set', 'callbacks', 'checkpoints'], varargs=None, keywords=None, defaults=(None, None, None, None)), ('document', 'c68fe1cb95d90762b57c309cae9b99d9'))
paddle.fluid.backward.gradients (ArgSpec(args=['targets', 'inputs', 'target_gradients', 'no_grad_set'], varargs=None, keywords=None, defaults=(None, None)), ('document', 'e2097e1e0ed84ae44951437bfe269a1b'))
paddle.fluid.regularizer.L1DecayRegularizer ('paddle.fluid.regularizer.L1DecayRegularizer', ('document', '34603757e70974d2fcc730643b382925'))
paddle.fluid.regularizer.L1DecayRegularizer.__init__ (ArgSpec(args=['self', 'regularization_coeff'], varargs=None, keywords=None, defaults=(0.0,)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
......@@ -1058,11 +1058,11 @@ paddle.fluid.LoDTensor.shape shape(self: paddle.fluid.core_avx.Tensor) -> List[i
paddle.fluid.LoDTensorArray ('paddle.fluid.core_avx.LoDTensorArray', ('document', 'e9895b67ba54438b9c0f7053e18966f5'))
paddle.fluid.LoDTensorArray.__init__ __init__(self: paddle.fluid.core_avx.LoDTensorArray) -> None
paddle.fluid.LoDTensorArray.append append(self: paddle.fluid.core_avx.LoDTensorArray, tensor: paddle.fluid.core_avx.LoDTensor) -> None
paddle.fluid.CPUPlace ('paddle.fluid.core_avx.CPUPlace', ('document', '6014005ef2649045b77d502aeb6cd7f9'))
paddle.fluid.CPUPlace ('paddle.fluid.core_avx.CPUPlace', ('document', 'd269ec68ce9b102ab10610e89ffa06e1'))
paddle.fluid.CPUPlace.__init__ __init__(self: paddle.fluid.core_avx.CPUPlace) -> None
paddle.fluid.CUDAPlace ('paddle.fluid.core_avx.CUDAPlace', ('document', '6a6cd8ed607beb951692c4b066d08c94'))
paddle.fluid.CUDAPlace ('paddle.fluid.core_avx.CUDAPlace', ('document', 'f862cb3e5596a3920102f1b1238c223b'))
paddle.fluid.CUDAPlace.__init__ __init__(self: paddle.fluid.core_avx.CUDAPlace, arg0: int) -> None
paddle.fluid.CUDAPinnedPlace ('paddle.fluid.core_avx.CUDAPinnedPlace', ('document', 'afd58ea5d390b5ea06ca70291a266d45'))
paddle.fluid.CUDAPinnedPlace ('paddle.fluid.core_avx.CUDAPinnedPlace', ('document', '1320ef739c81c95385330dab3fe6e80b'))
paddle.fluid.CUDAPinnedPlace.__init__ __init__(self: paddle.fluid.core_avx.CUDAPinnedPlace) -> None
paddle.fluid.ParamAttr ('paddle.fluid.param_attr.ParamAttr', ('document', '7b5bfe856689036b8fffb71af1558e5c'))
paddle.fluid.ParamAttr.__init__ (ArgSpec(args=['self', 'name', 'initializer', 'learning_rate', 'regularizer', 'trainable', 'gradient_clip', 'do_model_average'], varargs=None, keywords=None, defaults=(None, None, 1.0, None, True, None, True)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
......
paddle/fluid/pybind/pybind.cc
浏览文件 @ ce468817
......@@ -832,9 +832,23 @@ All parameter, weight, gradient are variables in Paddle.
py::class_<platform::Communicator>(m, "Communicator").def(py::init<>());
#endif
py::class_<platform::CUDAPlace>(m, "CUDAPlace", R"DOC(
CUDAPlace is a descriptor of a device. It represents a GPU, and each CUDAPlace
has a dev_id to indicate the number of cards represented by the current CUDAPlace.
**Note**:
For multi-card tasks, please use `FLAGS_selected_gpus` environment variable to set the visible GPU device.
The next version will fix the problem with `CUDA_VISIBLE_DEVICES` environment variable.
CUDAPlace is a descriptor of a device.
It represents a GPU device allocated or to be allocated with Tensor or LoDTensor.
Each CUDAPlace has a dev_id to indicate the graphics card ID represented by the current CUDAPlace,
staring from 0.
The memory of CUDAPlace with different dev_id is not accessible.
Numbering here refers to the logical ID of the visible graphics card, not the actual ID of the graphics card.
You can set visible GPU devices by setting the `CUDA_VISIBLE_DEVICES` environment variable.
When the program starts, visible GPU devices will be numbered from 0.
If `CUDA_VISIBLE_DEVICES` is not set, all devices are visible by default,
and the logical ID is the same as the actual ID.
Parameters:
id (int): GPU device ID.
Examples:
.. code-block:: python
......@@ -892,14 +906,14 @@ All parameter, weight, gradient are variables in Paddle.
.def("__str__", string::to_string<const platform::CUDAPlace &>);
py::class_<paddle::platform::CPUPlace>(m, "CPUPlace", R"DOC(
CPUPlace is a descriptor of a device. It represents a CPU, and the memory
CPUPlace can be accessed by CPU.
CPUPlace is a descriptor of a device.
It represents a CPU device allocated or to be allocated with Tensor or LoDTensor.
Examples:
.. code-block:: python
import paddle.fluid as fluid
cpu_place = fluid.CPUPlace()
cpu_place = fluid.CPUPlace()to be allocated
)DOC")
.def(py::init<>())
......@@ -912,8 +926,12 @@ All parameter, weight, gradient are variables in Paddle.
.def("__str__", string::to_string<const platform::CPUPlace &>);
py::class_<paddle::platform::CUDAPinnedPlace>(m, "CUDAPinnedPlace", R"DOC(
CUDAPinnedPlace is a descriptor of a device. The memory of CUDAPinnedPlace
can be accessed by GPU and CPU.
CUDAPinnedPlace is a descriptor of a device.
It refers to the page locked memory allocated by the CUDA function `cudaHostAlloc()` in the host memory.
The host operating system will not paging and exchanging the memory.
It can be accessed through direct memory access technology to speed up the copy of data between the host and GPU.
For more information on CUDA data transfer and `pinned memory`,
please refer to `official document <https://docs.nvidia.com/cuda/cuda-c-best-practices-guide/index.html#pinned-memory>`_ .
Examples:
.. code-block:: python
......
python/paddle/fluid/backward.py
浏览文件 @ ce468817
......@@ -919,29 +919,31 @@ def append_backward(loss,
callbacks=None,
checkpoints=None):
"""
Append backward part to main_program.
This function appends backward part to main_program.
A complete neural network training is made up of forward and backward
propagation. However, when we configure a network, we only need to
specify its forwrd part. The backward part is generated automatically
according to the forward part by this function.
specify its forward part. This function uses the chain rule to automatically
generate the backward part according to the forward part.
In most cases, users do not need to invoke this function manually. It
will be automatically invoked by the optimizer's `minimize` function.
In most cases, users do not need to invoke this function manually.
It will be automatically invoked by the optimizer's `minimize` function.
Args:
loss(Variable): The loss variable of the network.
parameter_list(list[string]|None): Names of parameters that need
Parameters:
loss( :ref:`api_guide_Variable_en` ): The loss variable of the network.
parameter_list(list of str, optional): Names of parameters that need
to be updated by optimizers.
If it is None, all parameters
will be updated.
Default: None
no_grad_set(set|None): Variables in the Block 0 whose gradients
Default: None.
no_grad_set(set of str, optional): Variable names in the :ref:`api_guide_Block_en` 0 whose gradients
should be ignored. All variables with
`stop_gradient=True` from all blocks will
be automatically added into this set.
Default: None
callbacks(list[callable object]|None): The callbacks are used for
If this parameter is not None, the names in this set will be added to the default set.
Default: None.
callbacks(list of callable object, optional): List of callback functions.
The callbacks are used for
doing some custom jobs during
backward part building. All
callable objects in it will
......@@ -950,23 +952,23 @@ def append_backward(loss,
into the program. The callable
object must has two input
parameters: 'block' and 'context'.
The 'block' is the block which
The 'block' is the :ref:`api_guide_Block_en` which
the new gradient operator will
be added to. The 'context' is a
map, whose keys are gradient
variable names and values are
corresponding original variables.
corresponding original :ref:`api_guide_Variable_en` .
In addition to this, the 'context'
has another special key-value pair:
the key is string '__current_op_desc__'
and the value is the op_desc of the
gradient operator who has just
triggered the callable object.
Default: None.
Returns:
list[(Variable,Variable)]: Pairs of parameter and its
corresponding gradients. The key is the parameter and the
value is gradient variable.
list of tuple ( :ref:`api_guide_Variable_en` , :ref:`api_guide_Variable_en` ): Pairs of parameter and its corresponding gradients.
The key is the parameter and the value is gradient variable.
Raises:
AssertionError: If `loss` is not an instance of Variable.
......@@ -974,17 +976,20 @@ def append_backward(loss,
Examples:
.. code-block:: python
# network configuration code
# loss from ...
import paddle.fluid as fluid
x = fluid.layers.data(name='x', shape=[13], dtype='float32')
y = fluid.layers.data(name='y', shape=[1], dtype='float32')
x = fluid.data(name='x', shape=[None, 13], dtype='float32')
y = fluid.data(name='y', shape=[None, 1], dtype='float32')
y_predict = fluid.layers.fc(input=x, size=1, act=None)
loss = fluid.layers.square_error_cost(input=y_predict, label=y)
avg_loss = fluid.layers.mean(loss)
param_grad_list = fluid.backward.append_backward(loss=avg_loss)
p_g_list1 = fluid.backward.append_backward(loss=avg_loss) # len(p_g_list1) == 2
p_g_list2 = fluid.backward.append_backward(loss=avg_loss, parameter_list=[p_g_list1[0][0].name]) # len(p_g_list1) == 1
p_g_list3 = fluid.backward.append_backward(loss=avg_loss, no_grad_set=set([p_g_list1[0][0].name])) # len(p_g_list1) == 1
p_g_list4 = fluid.backward.append_backward(loss=avg_loss, parameter_list=[p_g_list1[0][0].name], no_grad_set=set([p_g_list1[0][0].name])) # len(p_g_list1) == 0
"""
assert isinstance(loss, framework.Variable)
......
python/paddle/fluid/executor.py
浏览文件 @ ce468817
......@@ -67,11 +67,21 @@ def _switch_scope(scope):
@signature_safe_contextmanager
def scope_guard(scope):
"""
Change the global/default scope instance by Python `with` statement. All
variable in runtime will assigned to the new scope.
This function switches scope through python `with` statement.
Scope records the mapping between variable names and variables ( :ref:`api_guide_Variable` ),
similar to brackets in programming languages.
If this function is not invoked, all variables and variable names are recorded in the default global scope.
When users need to create variables with the same name,
they need to switch scopes through this function
if they do not want the mapping of variables with the same name to be overwritten.
After switching through the `with` statement,
all variables created in the `with` block will be assigned to a new scope.
Parameters:
scope: The new scope.
Args:
scope: The new global/default scope.
Returns:
None
Examples:
.. code-block:: python
......
python/paddle/fluid/framework.py
浏览文件 @ ce468817
......@@ -62,17 +62,20 @@ _dygraph_current_expected_place_ = None
def in_dygraph_mode():
"""
Check program status(tracer), Whether it runs in dygraph mode or not
This function checks whether the program runs in dynamic graph mode or not.
You can turn on dynamic graph mode with :ref:`api_fluid_dygraph_guard` api.
Returns:
out (boolean): True if the program is running in dynamic graph mode
bool: Whether the program is running in dynamic graph mode.
Examples:
.. code-block:: python
import paddle.fluid as fluid
if fluid.in_dygraph_mode():
pass
print('running in dygraph mode')
else:
print('not running in dygraph mode')
"""
return _dygraph_tracer_ is not None
......@@ -149,25 +152,29 @@ def is_compiled_with_cuda():
def cuda_places(device_ids=None):
"""
Create a list of :code:`fluid.CUDAPlace` objects.
**Note**:
For multi-card tasks, please use `FLAGS_selected_gpus` environment variable to set the visible GPU device.
The next version will fix the problem with `CUDA_VISIBLE_DEVICES` environment variable.
This function creates a list of :code:`fluid.CUDAPlace` objects.
If :code:`device_ids` is None, environment variable of
:code:`FLAGS_selected_gpus` would be checked first. If
:code:`FLAGS_selected_gpus` would be checked first. For example, if
:code:`FLAGS_selected_gpus=0,1,2`, the returned list would
be [fluid.CUDAPlace(0), fluid.CUDAPlace(1), fluid.CUDAPlace(2)].
If :code:`FLAGS_selected_gpus` is not set, all visible
gpu places would be returned.
gpu places would be returned according to the :code:`CUDA_VISIBLE_DEVICES` environment variable.
If :code:`device_ids` is not None, it should be the device
ids of gpus. For example, if :code:`device_ids=[0,1,2]`,
ids of GPUs. For example, if :code:`device_ids=[0,1,2]`,
the returned list would be
[fluid.CUDAPlace(0), fluid.CUDAPlace(1), fluid.CUDAPlace(2)].
Args:
device_ids (None|list(int)|tuple(int)): gpu device id list.
Parameters:
device_ids (list or tuple of int, optional): list of GPU device ids.
Returns:
out (list(fluid.CUDAPlace)): gpu place list.
list of fluid.CUDAPlace: Created GPU place list.
Examples:
.. code-block:: python
......@@ -187,18 +194,20 @@ def cuda_places(device_ids=None):
def cpu_places(device_count=None):
"""
Create a list of :code:`fluid.CPUPlace` objects.
This function creates a list of :code:`fluid.CPUPlace` objects, and returns the created list.
If :code:`device_count` is None, the device count would
be determined by environment variable :code:`CPU_NUM`.
If :code:`CPU_NUM` is not set, the default value is 1,
i.e. CPU_NUM=1.
:code:`CPU_NUM` indicates the number of devices used in the current task.
The running of the program can be accelerated if :code:`CPU_NUM` is the same as the number of physical cores.
Args:
device_count (None|int): device number.
Parameters:
device_count (int, optional): device number. Default: None.
Returns:
out (list(fluid.CPUPlace)): cpu place list.
list of fluid.CPUPlace: Created list of CPU places.
Examples:
.. code-block:: python
......@@ -214,18 +223,20 @@ def cpu_places(device_count=None):
def cuda_pinned_places(device_count=None):
"""
Create a list of :code:`fluid.CUDAPinnedPlace` objects.
This function creates a list of :code:`fluid.CUDAPinnedPlace` objects.
If :code:`device_count` is None, the device count would
be determined by environment variable :code:`CPU_NUM`.
If :code:`CPU_NUM` is not set, the device count would
be determined by :code:`multiprocessing.cpu_count()`.
If :code:`CPU_NUM` is not set, the default value is 1,
i.e. CPU_NUM=1.
:code:`CPU_NUM` indicates the number of devices used in the current task.
The running of the program can be accelerated if :code:`CPU_NUM` is the same as the number of physical cores.
Args:
device_count (None|int): device number.
Parameters:
device_count (int, optional): device number. Default: None.
Returns:
out (list(fluid.CUDAPinnedPlace)): cuda pinned place list.
list of fluid.CUDAPinnedPlace: Created list of CUDA pinned places.
Examples:
.. code-block:: python
......
Markdown is supported
0% 或 .
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 或