rms_norm_op

→TensorImplPtr

int rms_norm_op(const int & x, const int & gamma, double eps)

source edit

Python wrappers

pylucid.nn.functional.rms_normfunction

Public free-function entry point for RMS Normalization.

Thin wrapper that delegates to RMSNormBackward::forward. This is the symbol the pybind11 binding layer forwards to from lucid.nn.functional.rms_norm.

See Also

RMSNormBackward : Autograd node implementing the forward + backward. layer_norm_op : Variant that also subtracts the mean and adds a bias.

Parameters

xTensorImplPtr

Input tensor of any shape with rank >= rank(gamma).

gammaTensorImplPtr

Per-element scale; must match the trailing dims of x.

epsdouble

Numerical-stability constant (typically 1e-8).

Returns

TensorImplPtr

Normalised output, same shape as x.