Function dfdx::losses::cross_entropy_with_logits_loss

pub fn cross_entropy_with_logits_loss<S: Shape, E: Dtype, D: Device<E>, T: Tape<E, D>>(
    logits: Tensor<S, E, D, T>,
    target_probs: Tensor<S, E, D>
) -> Tensor<Rank0, E, D, T>

Cross entropy loss. This computes: -(logits.log_softmax() * target_probs).sum(-1).mean()

This will call log_softmax(logits), so make sure logits is not the output from softmax() or log_softmax() already.

Arguments

• logits: The un-normalized output from a model. log_softmax() is called in this function
• target_probs: Target containing probability vectors NOT class indices.