Returns the number of TestPartResult objects in the array.

Return number of total elements

Move the tensor data onto a given device. Args: device: a swig Device converted from CudaGPU or CppCPU or OpenclGPU

Copy the data from the numpy array. Args: np_array: source numpy array offset (int): destination offset

Return a new tensor with the given shape, and the original tensor is not changed. Args: shape (list<int>): new shape,

data type, including kFloat16, kFloat32, kInt

Return a new tensor with the given shape, and the original tensor is not changed. Args: shape (list<int>): new shape,

Gets the outcome of the test part.

Args: dy (CTensor): data for the `dL / dy`, L is the loss. Raises: AssertionError: no backward function for t

Return the programming language for this device.

Performs backward propagation from the loss and parameter update with sparsification. THIS IS A EXPERIMENTAL FUNCTION FOR RESEARCH PURPOSE:

Set all elements of the tensor to be the give value. Args: x (float): a float value to be set to all elements. inplac

Returns: a new Tensor which does deep copy of this tensor

parse the ONNX and to create layers Args: model (ModelProto): the loaded ONNX model device (string): CPU or C

Element-wise max of each of the input tensors (with Numpy-style broadcasting support). Args: *x (a list of Tensor): List of tensors f

Generate a value for each element following a Gaussian distribution. Args: mean (float): mean of the distribution std

Set all elements of the tensor to be the give value. Args: x (float): a float value to be set to all elements. inplac

Client push weights to server. Namely server pull weights from clients.

Signal and silence signaling NaN. \param nan half-precision NaN value \return quiet NaN \exception FE_INVALID if \a nan is signaling NaN

return new tensor

Performs backward propagation from the loss and parameter update using asynchronous training. THIS IS A EXPERIMENTAL FUNCTION FOR RESEARCH PU

Call Write method to write a parameter Args: param_name (string): name of the parameter param_val (Tensor): value ten

Performs all reduce of a tensor for distributed training. Args: tensor(Tensor): a tensor to be all-reduced

Exponential function. This function is exact to rounding for all rounding modes. See also:** Documentation for [std::exp](https://en.cppreference.com

Performs backward propagation and parameter update, with FP16 precision communication. THIS IS A EXPERIMENTAL FUNCTION FOR RESEARCH PURPOSE:

Returns: the number of elements of the tensor.

Element-wise min of each of the input tensors (with Numpy-style broadcasting support). Args: *x (a list of Tensor): List of tenso

Compile and initialize the model This function will automatically derive the shape of parameters in each sublayer based on the shape

====================Tensor Operations=======================================

Change the data type. Args: dtype: accepts 'int', 'float', 'singa.kFloat32', 'singa.kInt' Returns: new tenso

Call read method to load all (param_name, param_val) Returns: a dict of (parameter name, parameter Tensor)

Generate a value for each element following a uniform distribution. Args: low (float): the lower bound high (float):

Sets a new value, overriding the one supplied in the constructor.

To transpose the tensor Args: axes: axes to transpose Returns: new transposed tensor

this function calls singa global functions that returns Tensor and create new python Tensor instance e.g., Tensor [singa_func](args..

Each layer instance would optionally have a name. Used for debugging and logging.

Gets the text streamed to this object so far as an std::string. Each '\0' character in the buffer is replaced with "\\0".

this function calls singa global functions that returns Tensor and create new python Tensor instance e.g., Tensor [singa_func](args..

Copy the data from the numpy array or other Tensor instance Args: t (Tensor or np array): source Tensor or numpy array

Copy the data from the numpy array or other Tensor instance Args: t (Tensor or np array): source Tensor or numpy array

Start the client. This method will first connect to the server. Then global rank is sent to the server.

To transpose the tensor Args: axes: axes to transpose Returns: new transposed tensor

Raise domain error and return NaN. return quiet NaN \exception FE_INVALID

Raise floating-point exception. \param flags exceptions to raise \param cond condition to raise exceptions for

history = history * momentum + grad * lr value = value - history

Helpers for printing colored strings to stdout. Note that on Windows, we cannot simply emit special characters and have the terminal change colors. Th

Args: input_size (int): The number of expected features in the input x hidden_size (int): The number of features in

Todo(Wangwei) Get Block From The Memory manager