kaira.losses.text.LabelSmoothingLoss

Inheritance diagram for LabelSmoothingLoss

class kaira.losses.text.LabelSmoothingLoss(smoothing=0.1, classes=0, dim=-1)

Bases: BaseLoss

Label Smoothing Loss Module.

This module implements label smoothing to prevent overconfidence.

Methods

__init__	Initialize the LabelSmoothingLoss module.
forward	Forward pass through the LabelSmoothingLoss module.

__init__(smoothing=0.1, classes=0, dim=-1)

Initialize the LabelSmoothingLoss module.

Parameters:

• smoothing (float) – Smoothing factor. Default is 0.1.

• classes (int) – Number of classes. Default is 0.

• dim (int) – Dimension to reduce. Default is -1.

forward(x: Tensor, target: Tensor) → Tensor

Forward pass through the LabelSmoothingLoss module.

Parameters:

• x (torch.Tensor) – The input logits tensor.

• target (torch.Tensor) – The target tensor.

Returns:

The label smoothing loss.

Return type:

torch.Tensor