AirTransposeConvOptions

AirTransposeConvOptions()

Options for the TRANSPOSE_CONV operator.

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

padding_height: int = Field(0, description='Padding for top & bottom')

pixels to pad input top & bottom (default 0).

padding_width: int = Field(0, description='Padding for left & right')

pixels to pad input left & right (default 0).

padding_offsets_height: int = Field(0, description='Padding for top & bottom offsets')

Padding for top & bottom offsets (default 0).

padding_offsets_width: int = Field(0, description='Padding for left & right offsets')

Padding for left & right offsets (default 0).

stride_height: int = Field(1, description='Vertical stride')

vertical stride (default 1).

stride_width: int = Field(1, description='Horizontal stride')

horizontal stride (default 1).

dilation_height: int = Field(1, description='Vertical dilation')

vertical dilation factor (default 1).

dilation_width: int = Field(1, description='Horizontal dilation')

horizontal dilation factor (default 1).

activation: AirActivationType = Field(AirActivationType.NONE, description='Fused activation function')

fused activation (default NONE).

bias_tensor_type: AirTensorType = Field(AirTensorType.FLOAT32, description='Bias tensor type')

bias tensor type (default FLOAT32).

output_shape: list[int] | None = Field(None, description='Output shape from a constant output_shape input')

output shape from a constant output_shape input (default None).