SqueezeV3

English | 简体中文

Product Support Status

Product	Supported
Ascend 950PR&950DT products	√
Atlas A3 products	√
Atlas A2 products	√
Atlas 200I/500 A2 inference products	√
Atlas inference products	√
Atlas training products	√

Function Description

API function: removes dimensions whose size is 1 from a tensor. axes is provided as an optional input. Example:

If axes is empty, all dimensions whose size is 1 are removed:

>>> x = torch.randn(2, 1, 3)
>>> torch.squeeze(x).shape
(2, 3)

If axes is specified, only the specified dimensions whose size is 1 are removed:

>>> x = torch.randn(2, 1, 3)
>>> torch.squeeze(x, 1).shape
(2, 3)

Parameter Description

Parameter Name	Input/Output/Attribute	Description	Data Type	Data Format
x	Input	Input tensor to be squeezed	All types	ND
axes	Optional input	List of dimensions to be removed. The value ranges from [-rank(x), rank(x)) If it is not specified or is empty, all dimensions whose size is 1 are removed If it is specified, only the specified dimensions are removed, and the size of each specified dimension must be 1	INT64	ND
y	Output	Squeezed output tensor, whose data type is the same as that of the input x	All types	ND

Constraint Description

• If axes is specified, the size of the dimension corresponding to axes must be 1 or -1. Otherwise, the operator fails to execute.
• If the value of axes is out of the range [-rank(x), rank(x)), the operator fails to execute.
• If the input has a dynamic shape (containing -1) and axes is empty, whether the -1 dimension is 1 cannot be determined, and the output shape is set to -2.
• If the input shape is -2, the output shape is -2.
• If the input shape is determined but axes cannot be obtained at the compilation stage (for example, axes comes from an upstream computation result), the output shape is set to -2.

Invocation Description

Invocation Method	Sample Code	Description
Graph-mode invocation	test_geir_squeezev3	Invokes the SqueezeV3 operator by building a graph through the operator IR.