mindspore.Parameter

class mindspore.Parameter(default_input, name=None, requires_grad=True, layerwise_parallel=False, parallel_optimizer=True, storage_format='')

Parameter is a Tensor subclass, when they are assigned as Cell attributes they are automatically added to the list of its parameters, and will appear, e.g. in cell.get_parameters() iterator.

Note

• In auto_parallel mode of SEMI_AUTO_PARALLEL and AUTO_PARALLEL, if init Parameter by a Tensor, the type of Parameter will be Tensor. Tensor will save the shape and type info of a tensor with no memory usage.

• The shape can be changed while compiling for auto-parallel. Call init_data will return a Tensor Parameter with initialized data.

• If there is an operator in the network that requires part of the inputs to be Parameter, then the Parameters as this part of the inputs are not allowed to be cast.

• Give each Parameter a unique name to facilitate subsequent operations and updates. If there are two or more Parameter objects with the same name in a network, will be prompted to set a unique name when defining.

• When directly printing a Parameter, you cannot view the actual values contained inside it. You need to use the Parameter.asnumpy() method to access the actual values.

Parameters

• default_input (Union[Tensor, int, float, numpy.ndarray, list]) – Parameter data, to initialize the parameter data.

• name (str) –

  Name of the parameter. Default: None . If two or more Parameter objects with the same name exist in a network, you will be prompted to set a unique name when defining them.

  1) If the parameter is not given a name, the default name is its variable name. For example, the name of param_a below is name_a, and the name of param_b is the variable name param_b.

  self.param_a = Parameter(Tensor([1], ms.float32), name="name_a")
  self.param_b = Parameter(Tensor([2], ms.float32))
  

  2) If parameter in list or tuple is not given a name, will give it a unique name. For example, the names of parameters below are Parameter$1 and Parameter$2.

  self.param_list = [Parameter(Tensor([3], ms.float32)),
                     Parameter(Tensor([4], ms.float32))]
  

  3) If the parameter is given a name, and the same name exists between different parameters, an exception will be thrown. For example, "its name 'name_a' already exists." will be thrown.

  self.param_a = Parameter(Tensor([1], ms.float32), name="name_a")
  self.param_tuple = (Parameter(Tensor([5], ms.float32), name="name_a"),
                      Parameter(Tensor([6], ms.float32)))
  

  4) If a parameter appear multiple times in list or tuple, check the name of the object only once. For example, the following example will not throw an exception.

  self.param_a = Parameter(Tensor([1], ms.float32), name="name_a")
  self.param_tuple = (self.param_a, self.param_a)
  

• requires_grad (bool) – True if the parameter requires gradient. Default: True .

• layerwise_parallel (bool) – When layerwise_parallel is true in data/hybrid parallel mode, broadcast and gradients communication would not be applied to the Parameter. Default: False .

• parallel_optimizer (bool) – It is used to filter the weight shard operation in SEMI_AUTO_PARALLEL or AUTO_PARALLEL mode. It works only when enable parallel optimizer in mindspore.set_auto_parallel_context(). Default: True .

• storage_format (str) – Only Ascend device target is supported. It is used to specify the format of the weight loaded to the device. By default, the format is not changed. The optional values are "FRACTAL_NZ" , "NC1HWC0" , "FRACTAL_Z" , etc. Default: "" .

Examples

>>> import numpy as np
>>> import mindspore
>>> from mindspore import Parameter, Tensor, ops, nn
>>>
>>> class Net(nn.Cell):
...     def __init__(self):
...         super(Net, self).__init__()
...         self.matmul = ops.MatMul()
...         self.weight = Parameter(Tensor(np.ones((1, 2)), mindspore.float32), name="w", requires_grad=True)
...
...     def construct(self, x):
...         out = self.matmul(self.weight, x)
...         return out
>>> net = Net()
>>> x = Tensor(np.ones((2, 1)), mindspore.float32)
>>> print(net(x))
[[2.]]
>>> net.weight.set_data(Tensor(np.zeros((1, 2)), mindspore.float32))
>>> print(net(x))
[[0.]]

add_pipeline_stage(stage)

Add a pipeline stage to the parameter.

Parameters

stage (int) – The pipeline stage to be added.

Raises

TypeError – If stage is not a positive number or not int type.

property cache_enable

Return whether the parameter is cache enable.

Examples

>>> from mindspore import Tensor, Parameter
>>> import numpy as np
>>> x = Parameter(Tensor(np.array([1, 2], dtype=np.float32)), name="param")
>>> x.cache_enable=True
>>> x.cache_enable
True

property cache_shape

Return the cache shape corresponding to the parameter if use cache.

Examples

>>> from mindspore import Tensor, Parameter
>>> import numpy as np
>>> x = Parameter(Tensor(np.array([1, 2], dtype=np.float32)), name="param")
>>> x.cache_enable=True
>>> x.cache_shape=[1, 2]
>>> x.cache_shape
[1, 2]

clone(init='same')

Clone the parameter.

Parameters

init (Union[Tensor, str, numbers.Number]) – Initialize the shape and dtype of the parameter. If init is a Tensor or numbers.Number, clone a new parameter with the same shape and dtype, and the data of the new parameter will be set according to init. If init is a str, the init should be the alias of the class inheriting from Initializer. For example, if init is 'same', clone a new parameter with the same data, shape, and dtype. Default: 'same'.

Returns

Parameter, a new parameter.

Examples

>>> from mindspore import Tensor, Parameter
>>> import numpy as np
>>> x = Parameter(Tensor(np.array([1, 2], dtype=np.float32)), name="param")
>>> y = x.clone()

property comm_fusion

Get the fusion type (int) for communication operators corresponding to this parameter.

In AUTO_PARALLEL and SEMI_AUTO_PARALLEL mode, some communication operators used for parameters or gradients aggregation are inserted automatically. The value of comm_fusion must be greater than or equal to 0. When the value of comm_fusion is 0 , operators will not be fused together.

Examples

>>> from mindspore import Tensor, Parameter
>>> import numpy as np
>>> x = Parameter(Tensor(np.array([1, 2], dtype=np.float32)), name="param")
>>> x.comm_fusion = 3
>>> x.comm_fusion
3

copy()

Copy the parameter.

Returns

Parameter, a new parameter.

Examples

>>> from mindspore import Tensor, Parameter
>>> import numpy as np
>>> x = Parameter(Tensor(np.array([1, 2], dtype=np.float32)), name="param")
>>> y = x.copy()

property data

Return the parameter object.

Examples

>>> from mindspore import Tensor, Parameter
>>> import numpy as np
>>> x = Parameter(Tensor(np.array([[1, 2], [3, 4]], dtype=np.float32)), name="param")
>>> x.data
Parameter (name=param, shape=(2, 2), dtype=Float32, requires_grad=True)

init_data(layout=None, set_sliced=False)

Initialize the parameter's data.

Parameters

• layout (Union[None, tuple]) –

  The parameter's layout info. layout [dev_mat, tensor_map, slice_shape, filed_size, uniform_split, opt_shard_group]. Default: None. It's not None only in 'SEMI_AUTO_PARALLEL' or 'AUTO_PARALLEL' mode.

  • dev_mat (list(int)): The parameter's device matrix.

  • tensor_map (list(int)): The parameter's tensor map.

  • slice_shape (list(int)): The parameter's slice shape.

  • filed_size (int): The parameter's filed size.

  • uniform_split (bool): Whether the parameter is split evenly.

  • opt_shard_group (str): The group of the parameter while running optimizer parallel.

• set_sliced (bool) – True if the parameter is set sliced after initializing the data. Default: False.

Returns

Parameter, the Parameter after initializing data. If current Parameter was already initialized before, returns the same initialized Parameter.

Raises

• RuntimeError – If it is from Initializer, and parallel mode has changed after the Initializer created.

• ValueError – If the length of the layout is less than 6.

• TypeError – If layout is not tuple.

Examples

>>> from mindspore import Tensor, Parameter
>>> import numpy as np
>>> x = Parameter(Tensor(np.array([[1, 2], [3, 4]], dtype=np.float32)), name="param")
>>> x.init_data()

property inited_param

Get the new parameter after call the init_data.

Default is a None, If self is a Parameter without data, after call the init_data the initialized Parameter with data will be recorded here.

Examples

>>> from mindspore import Tensor, Parameter
>>> import numpy as np
>>> x = Parameter(Tensor(np.array([1, 2], dtype=np.float32)), name="param")
>>> x.inited_param

property key

Return the parameter unique key.

Examples

>>> from mindspore import Tensor, Parameter
>>> import numpy as np
>>> x = Parameter(Tensor(np.array([1, 2], dtype=np.float32)), name="param")
>>> x.key = 2
>>> x.key
2

property layerwise_parallel

Get the layerwise parallel status(bool) of the parameter.

When layerwise_parallel is True in DATA_PARALLEL and HYBRID_PARALLEL parallel mode, broadcast and gradients communication would not be applied to parameters.

Examples

>>> from mindspore import Tensor, Parameter
>>> import numpy as np
>>> x = Parameter(Tensor(np.array([1, 2], dtype=np.float32)), name="param")
>>> x.layerwise_parallel = True
>>> x.layerwise_parallel
True

property name

Get the name of the parameter.

Examples

>>> from mindspore import Tensor, Parameter
>>> import numpy as np
>>> x = Parameter(Tensor(np.array([1, 2], dtype=np.float32)), name="param")
>>> x.name = "param1"
>>> x.name
'param1'

property parallel_optimizer

Get the optimizer parallel status(bool) of the parameter.

It is used to filter the weight shard operation in AUTO_PARALLEL and SEMI_AUTO_PARALLEL mode. It works only when enable parallel optimizer in mindspore.set_auto_parallel_context().

Examples

>>> from mindspore import Tensor, Parameter
>>> import numpy as np
>>> x = Parameter(Tensor(np.array([1, 2], dtype=np.float32)), name="param")
>>> x.parallel_optimizer = True
>>> x.parallel_optimizer
True

property parallel_optimizer_comm_recompute

Get the communication recompute status(bool) of optimizer parallel for the parameter.

In AUTO_PARALLEL and SEMI_AUTO_PARALLEL mode, when applying parallel optimizer, some mindspore.ops.AllGather operators used for parameters gathering are inserted automatically. It is used to control the recompute attr for those mindspore.ops.AllGather operators.

Note

• Only Graph mode is supported.

• It is recommended to use cell.recompute(parallel_optimizer_comm_recompute=True/False) to configure the AllGather operators introducing by parallel optimizer rather than using this interface directly.

Examples

>>> from mindspore import Tensor, Parameter
>>> import numpy as np
>>> x = Parameter(Tensor(np.array([1, 2], dtype=np.float32)), name="param")
>>> x.parallel_optimizer_comm_recompute = True
>>> x.parallel_optimizer_comm_recompute
True

property requires_grad

Return whether the parameter requires gradient.

Examples

>>> from mindspore import Tensor, Parameter
>>> import numpy as np
>>> x = Parameter(Tensor(np.array([1, 2], dtype=np.float32)), name="param")
>>> x.requires_grad = True
>>> x.requires_grad
True

set_data(data, slice_shape=False)

Set Parameter's data.

Parameters

• data (Union[Tensor, int, float]) – New data.

• slice_shape (bool) – If slice the parameter is set to True, the shape consistency will not be checked. Default: False. When slice_shape is True, and the shapes are not consistent, a ValueError will be thrown.

Returns

Parameter, the parameter after set data.

Examples

>>> from mindspore import Tensor, Parameter
>>> import numpy as np
>>> x = Parameter(Tensor(np.array([[1, 2], [3, 4]], dtype=np.float32)), name="param")
>>> x.set_data(Tensor(np.array([[6, 6], [6, 6]], dtype=np.float32)))
Parameter (name=param, shape=(2, 2), dtype=Float32, requires_grad=True)

set_param_ps(init_in_server=False)

Set whether the trainable parameter is updated by parameter server and whether the trainable parameter is initialized on server.

Note

It only works when a running task is in the parameter server mode. It is supported only in graph mode.

Parameters

init_in_server (bool) – Whether trainable parameter updated by parameter server is initialized on server. Default: False.

Tutorial Examples:

• Parameter Server Mode

property sliced

Get slice status of the parameter.

Examples

>>> from mindspore import Tensor, Parameter
>>> import numpy as np
>>> x = Parameter(Tensor(np.array([1, 2], dtype=np.float32)), name="param")
>>> x.sliced = True
>>> x.sliced
True

property unique

Whether the parameter is already unique or not.

Examples

>>> from mindspore import Tensor, Parameter
>>> import numpy as np
>>> x = Parameter(Tensor(np.array([1, 2], dtype=np.float32)), name="param")
>>> x.unique = True
>>> x.unique
True

value()

Return the value of parameter object.

Examples

>>> from mindspore import Tensor, Parameter
>>> import numpy as np
>>> x = Parameter(Tensor(np.array([1, 2], dtype=np.float32)), name="param")
>>> x_value = x.value()
>>> print(x_value)
[1.  2.]