Implement softplus

[BruceDai]

@fdwr @huningxin PTAL, thanks.

[fdwr]

Oh, I hadn't noticed until now that WebNN's softplus is like PyTorch but differs from ONNX's.

WebNN: ln(exp(steepness * x) + 1) / steepness.
ONNX: ln(exp(x) + 1)
PyTorch: log(exp(scale ∗ x)+1) / scale

[huningxin]

Oh, I hadn't noticed until now that WebNN's softplus is like PyTorch but differs from ONNX's.

WebNN: ln(exp(steepness * x) + 1) / steepness. ONNX: ln(exp(x) + 1) PyTorch: log(exp(scale ∗ x)+1) / scale

@fdwr , thanks for the good catching. Do you see any issues here?

[fdwr]

Do you see any issues here?

@huningxin : No issue, just an observation, as WebNN's softplus is directly compatible with PyTorch and a superset of ONNX. If was a subset, then there would be more of an issue.