Skip to main content

QConv

QConv(in_channels, out_channels, kernel, *, stride=None, padding=None, dilation=None, groups=1, bias=True, padding_mode="zeros", value=None, activation=None, dtype=None, device=None)

Static-quant channels-last convolution: quantized weight AND quantized activations; the forward returns floating point, or a quantized tensor once an output requant is bound with :meth:set_output_quantization. The same knobs as :class:Conv.

__init__​

__init__(self, in_channels: 'int', out_channels: 'int', kernel: 'Sequence[int]', *, stride: 'Sequence[int] | None' = None, padding: 'Sequence[int] | None' = None, dilation: 'Sequence[int] | None' = None, groups: 'int' = 1, bias: 'bool' = True, padding_mode: 'PaddingModeLike' = 'zeros', value: 'float | None' = None, activation: 'ActivationLike' = None, dtype: 'DtypeLike | None' = None, device: 'Device | str | None' = None) -> 'None'

Initialize self. See help(type(self)) for accurate signature.

forward​

forward(self, input: 'QTensor') -> 'Tensor'

forward(input) -> Tensor

Quantized channels-last activations to [batch, *spatial', out_channels] (floating point, or quantized once an output requant is bound).

set_output_quantization​

set_output_quantization(self, scale: 'Tensor', zero_point: 'Tensor | None' = None, *, quant_axis: 'int' = -1, out_dtype: 'DtypeLike | None' = None) -> 'None'

set_output_quantization(scale, zero_point=None, *, quant_axis=-1, out_dtype=None) -> None

Bind the output requant: the forward then returns a quantized tensor at the bound scheme. Never calling it keeps the floating-point output path.