This document is relevant for: Inf2, Trn1, Trn1n

nki.language.rms_norm#

nki.language.rms_norm(x, w, axis, n, epsilon=1e-06, dtype=None, compute_dtype=None, mask=None, **kwargs)[source]#

Apply Root Mean Square Layer Normalization.

Parameters:

  • x – input tile

  • w – weight tile

  • axis – axis along which to compute the root mean square (rms) value

  • n – total number of values to calculate rms

  • epsilon – epsilon value used by rms calculation to avoid divide-by-zero

  • dtype – (optional) data type to cast the output type to (see Supported Data Types for more information); if not specified, it will default to be the same as the data type of the input tile.

  • compute_dtype – (optional) dtype for the internal computation - currently `dtype` and `compute_dtype` behave the same, both sets internal compute and return dtype.

  • mask – (optional) a compile-time constant predicate that controls whether/how this instruction is executed (see NKI API Masking for details)

Returns:

`` x / RMS(x) * w ``

This document is relevant for: Inf2, Trn1, Trn1n