exp_op

→TensorImplPtr

int exp_op(const int & a)

Compute $y = e^x$ element-wise. Allocates a fresh output of the same shape and dtype as a and delegates to ExpBackward::forward.

See Also

ExpBackward — backward node.

aTensorImplPtr

Input tensor of any shape. Half-precision inputs are upcast to F32 before computation, per AmpPolicy::ForceFP32.

TensorImplPtr

Output tensor of the same shape and dtype. Large positive values overflow to +inf; large negative values flush to 0.

Loading…