Module: Toy::LLM::Primitives::RMSNorm

Defined in:
lib/toy/llm/primitives/rms_norm.rb,
lib/toy/llm/primitives/rms_norm_cuda.rb,
lib/toy/llm/primitives/rms_norm_metal.rb

Constant Summary collapse

NAME = 
:rms_norm

Class Method Summary collapse

Class Method Details

.build(sess, x, gamma, eps) ⇒ Object

Root-mean-square normalisation: rms_norm(x) * gamma. gamma is the caller-owned scale tensor (block-owned param); eps is the Float epsilon added inside the rms. The primitive only consumes the gamma handle — it never allocates or frees it. Returns the normalised tensor handle.



33
34
35
 
# File 'lib/toy/llm/primitives/rms_norm.rb', line 33

def self.build(sess, x, gamma, eps)
  TinyNN.tnn_rms_norm(sess, x, gamma, eps)
end