QuantLowerTriangularLinear

Bases: QuantBase

Quantized :class:LowerTriangularLinear.

• out <= in: bottom out rows of the in x in lower-triangular matrix (block flush against the right edge of the weight matrix).
• out > in: leading in rows form a square lower-triangular block; trailing out - in rows are fully dense.

Fake-quantization is applied to weight * mask, so masked-out entries remain exactly zero through quantization.

Parameters:

Name	Type	Description	Default
in_features	int	Number of input features.	required
out_features	int	Number of output features.	required
bias	bool	Whether to include an additive bias.	True
act_func	str | None	Either "relu" or None.	None
ema_constant	float	EMA smoothing factor for observers.	0.01
device	str	Torch device for the parameters.	'cpu'