AirStridedSliceOptions

AirStridedSliceOptions()

Options for STRIDED_SLICE operator.

Provisional: importable, but outside the compatibility promise; a minor release can change or remove it.

begin: tuple[int, ...] = Field(..., description='Begin indices for each dimension')

Begin indices for the specified leading axes.

end: tuple[int, ...] = Field(..., description='End indices for each dimension')

End indices for the specified leading axes.

stride: tuple[int, ...] = Field(..., description='Stride values for each dimension')

Strides for the specified leading axes.

begin_mask: int = Field(0, description='Mask for begin indices')

Mask for begin indices (default 0).

end_mask: int = Field(0, description='Mask for end indices')

Mask for end indices (default 0).

ellipsis_mask: int = Field(0, description='Mask for ellipsis')

Mask for ellipsis (default 0).

new_axis_mask: int = Field(0, description='Mask for new axes')

Mask for new axes (default 0).

shrink_axis_mask: int = Field(0, description='Mask for shrink axes')

Mask for shrink axes (default 0).

offset: bool = Field(False, description='If True, apply offset to the begin and end indices')

If True, apply offset to the begin and end indices (default False).

get_slices(input_shape: tuple[int, ...], pad_to: int = 4) -> list[AirStrideSlice]

Compute each axis’s slice, then pad to the kernel’s pad_to dimensions.

Missing trailing axes are taken whole, as in LiteRT. The parameter vectors must have equal lengths and cannot exceed the input rank.

Parameters

Parameters of get_slices
Name	Type	Default	Description
`input_shape`	`tuple[int, ...]`	Required	The shape of the input tensor.
`pad_to`	`int`	4	The number of dimensions to pad the input shape to (default 4).

Returns

Returns of get_slices
Type	Description
`list[AirStrideSlice]`	A list of AirStrideSlice, each containing (begin, end, stride) for each axis.

Raises

Errors raised by get_slices
Type	Description
`UnsupportedModelError`	If ``ellipsis_mask`` or ``new_axis_mask`` is set.
`ConfigValueError`	If the parameter vectors differ in length or exceed the input rank.