Module loss 

Loss functions.

These are small, allocation-free helpers intended to be used like:

• run model.forward(...)
• compute d_output via a loss (e.g. mse_backward)
• run model.backward(...)
• update parameters with an optimizer

Enums

Loss

Supported loss functions.

Functions

bce_with_logits

Binary cross-entropy loss with logits.

bce_with_logits_backward

BCE-with-logits loss + gradient w.r.t logits.

mae

Mean absolute error (MAE) loss.

mae_backward

MAE loss + gradient w.r.t pred.

mse

Mean squared error (MSE) loss.

mse_backward

MSE loss + gradient w.r.t. pred.

softmax_cross_entropy

Softmax cross-entropy over a single sample.

softmax_cross_entropy_backward

Softmax cross-entropy + gradient w.r.t logits.