class

SinhBackward

SinhBackward

source edit

Python wrappers

pylucid.ops.sinhfunction

Autograd node for element-wise hyperbolic sine $y = \sinh(x)$ .

Saves the input x so the backward pass can evaluate $\cosh(x)$ and scale the upstream gradient by it. Defined on all of $\mathbb{R}$ with the identity $\sinh(x) = (e^x - e^{-x})/2$ .

Math

\frac{\partial y}{\partial x} = \cosh(x), \qquad \frac{\partial \mathcal{L}}{\partial x} = \cosh(x)\,\frac{\partial \mathcal{L}}{\partial y}.$$

Attributes

schema_v1OpSchema

Registered as "sinh" with AmpPolicy::Promote.

Notes

Dispatch: Accelerate vvsinhf/vvsinh (CPU) / MLX sinh (GPU). Output magnitude grows exponentially; large-magnitude inputs can overflow to +inf / -inf.

Static methods

static

dispatch

int dispatch(backend::IBackend & be, const int & a, const int & s, Dtype dt)

source edit

Forward — calls IBackend::sinh to compute $y = \sinh x$ .

Methods

grad_formula

int grad_formula(const int & g)

source edit

Backward — $\partial y/\partial x = \cosh x$ , scaled by grad_out.

class

SinhBackward

SinhBackward

source edit

Python wrappers

pylucid.ops.sinhfunction

Autograd node for element-wise hyperbolic sine $y = \sinh(x)$ .

Saves the input x so the backward pass can evaluate $\cosh(x)$ and scale the upstream gradient by it. Defined on all of $\mathbb{R}$ with the identity $\sinh(x) = (e^x - e^{-x})/2$ .

Math

\frac{\partial y}{\partial x} = \cosh(x), \qquad \frac{\partial \mathcal{L}}{\partial x} = \cosh(x)\,\frac{\partial \mathcal{L}}{\partial y}.$$

Attributes

schema_v1OpSchema

Registered as "sinh" with AmpPolicy::Promote.

Notes

Dispatch: Accelerate vvsinhf/vvsinh (CPU) / MLX sinh (GPU). Output magnitude grows exponentially; large-magnitude inputs can overflow to +inf / -inf.

Static methods

static

dispatch

int dispatch(backend::IBackend & be, const int & a, const int & s, Dtype dt)

source edit

Forward — calls IBackend::sinh to compute $y = \sinh x$ .

Methods

grad_formula

int grad_formula(const int & g)

source edit

Backward — $\partial y/\partial x = \cosh x$ , scaled by grad_out.