eazygrad.bce_with_logits_loss¶

eazygrad.bce_with_logits_loss(logits: _Tensor, target: _Tensor) → _Tensor[source]¶

Compute binary cross-entropy loss from unnormalized logits.

Parameters:

logits (_Tensor) – Input logits.
target (_Tensor) – Target tensor with values typically in [0, 1] and the same shape as logits.

Returns:

_Tensor – Scalar loss tensor.

Notes

This implementation uses a numerically stable formulation and internally averages the per-element loss.