QuantConv2d

Bases: QuantBase

Int8-quantized 2-D convolution with optional BatchNorm and ReLU.

When bn=True a :class:torch.nn.BatchNorm2d is applied after the convolution during training; for schema export BN is folded into the weight and bias.

Parameters:

Name	Type	Description	Default
in_channels	int	Number of input channels.	required
out_channels	int	Number of output channels.	required
kernel_size	int | tuple[int, int]	Kernel size.	required
stride	int | tuple[int, int]	Convolution stride.	1
padding	int | tuple[int, int] | tuple[int, int, int, int]	Zero-padding. int or 2-tuple (pad_h, pad_w) or 4-tuple (L, R, T, B).	0
bias	bool	Whether to include an additive bias.	True
device	str	Torch device for the underlying parameters.	'cpu'
act_func	str | None	Either "relu" or None.	None
ema_constant	float	EMA smoothing factor for observers.	0.01
bn	bool	If True, append a BatchNorm2d layer and fold it into weights for schema export.	False