UpperTriangularLinear

Bases: BaseModule

Linear layer with an upper-triangular weight, optionally extended with dense rows.

Designed to support in-place inference with a buffer of size max(in_features, out_features):

• When out <= in the weight is pure upper triangular.
• When out > in the first in rows are upper triangular and the remaining out - in rows are fully dense.

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 term.	True
device	str	Torch device for the parameters.	'cpu'
act_func	str | None	Either "relu" or None.	None