rpow_scalar_op

→TensorImplPtr

int rpow_scalar_op(double base, const int & a)

Element-wise scalar-base reverse power — returns $\text{base}^{a}$ with a fully wired autograd node.

Math

\frac{\partial y_i}{\partial x_i} = \ln(\text{base}) \cdot y_i$$ **Shape** Output shape equals input shape (elementwise). **See Also** `pow_scalar_op`, `RPowScalarBackward`.

basedouble

Scalar base (must be positive for a real-valued gradient).

aTensorImplPtr

Input tensor — the exponent (promoted to FP32 by the op schema).

TensorImplPtr

Output tensor with the same shape as a; gradient is $\ln(\text{base}) \cdot y \cdot g$ .

Gradient is undefined / non-real for ``base <= 0``; callers should

guard the base if they intend to backpropagate.