public override differentiate(inputs as NDArray<T>, upstream as NDArray<T>) as LayerGradients<T>
Computes a Jacobian-vector product without materializing a Jacobian.
| Parameter | Description |
|---|---|
inputs | Original logits. |
upstream | Same-shaped derivative of the probabilities. |
Input derivatives and no parameter derivatives.
dx = activation.differentiate(logits, upstream).inputs