xma.functional.cross_entropy

cross_entropy(x: Tensor, labels: Tensor, reduction: str = 'mean', logits_multiplier: float | None = None, *, kernel_backend: KernelBackend | None = None) → Tensor[source]

cross entropy loss

Parameters:

  • x (torch.Tensor) – logits

  • labels (torch.Tensor) – labels

  • reduction (str) – reduction method: “sum”, “mean” or None

  • logits_multiplier (float | None) – logits multiplier pre-multiplies logits, None implies 1. Defaults to None.

  • kernel_backend (KernelBackend | None) – KernelBackend

Returns:

loss

Return type:

Tensor