pypto.full

Applicable Products

Ascend 950PR/Ascend 950DT: Supported
Atlas A3 training products/Atlas A3 inference products: Supported
Atlas A2 training products/Atlas A2 inference products: Supported

Description

Creates a tensor of the specified size, filled with the given fill_value. The data type of the tensor is determined by dtype.

Prototype

full(size: List[int], fill_value: Union[int, float, Element], dtype: DataType, *, valid_shape: Optional[Union[List[int], List[SymbolicScalar]]] = None ) -> Tensor

Parameters

Parameter	Input/Output	Description
size	Input	Source operand that defines the shape of the output tensor. Supported data type: List[int]. Supported dimension range: 1 to 4 dimensions.
fill_value	Input	Source operand used to fill the values of the output tensor. Supported data types: int, float, and Element. When the value is of type int or float, it is automatically converted to the Element type, where int corresponds to DT_INT32 and float corresponds to DT_FP32. When other data types are required, they can be constructed through Element. Data types supported by Element: DT_FP32, DT_FP16, DT_BF16, DT_INT8, DT_INT16, DT_INT32, DT_UINT8, DT_UINT16, DT_UINT32, and DT_BOOL. The input must be of the same type as dtype; implicit conversion is not supported.
dtype	Input	Source operand that defines the type of the output tensor. Supported data types: DT_FP32, DT_FP16, DT_BF16, DT_INT8, DT_INT16, DT_INT32, DT_UINT8, DT_UINT16, DT_UINT32, and DT_BOOL. The input must be of the same type as fill_value; implicit conversion is not supported.
valid_shape	Input	Source operand that defines the dynamic shape of the output tensor. It is a keyword parameter used in dynamic graphs and can be omitted in static graphs. Supported types: List[SymbolicScalar] and List[int].

Return Value

Returns an output tensor. The data type of the tensor matches dtype, its shape is determined by size, and all elements are set to fill_value.

Constraints

valid_shape is used in dynamic graph scenarios.

In a dynamic graph scenario, if a [5,5] tensor needs to be generated with the ViewShape set to [2,2], the framework generates [2,2] tiles through the pypto.loop loop and concatenates them by offset. In this case, if valid_shape is not passed, the code generates a tensor filled entirely with [2,2] by default (for example, pypto.full([2,2], 1, pypto.DT_INT32)).

However, when the total size [5,5] is not divisible by the tile size [2,2], the valid shape of the tail tile (for example, [1,1]) cannot be automatically inferred by the framework. For example, the last row/column may contain only one element instead of a complete [2,2] tile. In this case, the actual valid shape of the tail tile must be explicitly specified through valid_shape, as follows:

pypto.full([2, 2], 1, pypto.DT_INT32, valid_shape=[pypto.min(2, 5 - 2 * b_idx), pypto.min(2, 5 - 2 * s_idx)]), where b_idx and s_idx represent the loop indices.
The dimension range of size is 1 to 4 dimensions, that is, the length range of size is [1, 4].

Examples

TileShape Setting Example

Before calling this operation API, set the TileShape through set_vec_tile_shapes.

The dimensions of the TileShape must be the same as those of the output.

For example, if the input size is [m, n], the output is [m, n], and the TileShape is set to [m1, n1], then m1 and n1 are used to tile the m and n axes, respectively.

pypto.set_vec_tile_shapes(4, 16)

API Call Example

# Valid shapes use keyword argument
x1 = 1.0 # must be 1.0; implicit conversion is not supported
y1 = pypto.full([2,2], x1, pypto.DT_FP32, valid_shape = [pypto.symbolic_scalar(2), pypto.symbolic_scalar(2)])

x2 = pypto.Element(pypto.DT_INT32,1)
y2 = pypto.full([2,2], x2, pypto.DT_INT32, valid_shape = [pypto.symbolic_scalar(2), pypto.symbolic_scalar(2)])

# In static graphs, validshape can be ignored
x3 = pypto.Element(pypto.DT_INT32,1)
y3 = pypto.full([2,2], x3, pypto.DT_INT32)

The results are as follows:

y1 output data: [[1.0,1.0], [1.0,1.0]]
y2 output data: [[1,1], [1,1]]
y3 output data: [[1,1], [1,1]]