SwiGLU applied to precomputed gate and linear vectors

Read the equation

The displayed full expression describes two affine projections, one passed through Swish, followed by elementwise multiplication. This function accepts already computed gate and linear vectors and multiplies gate·sigmoid(gate) by linear. Its x argument is used to set and validate length; the helper does not itself calculate the W, V, b, or c projections.

Terms & definitions

How it works

Read the definition from its right-hand expression to the quantity on the left. Its surrounding source records the intended input, state, or parameter context. Where the source leaves an operation unspecified, this page preserves that gap.

Assumptions & limits

Source recovery establishes that the expression appears in this record. Its mathematical validity, implementation behavior, and interpretation must be assessed separately. See the linked source and research notes for the checks available for this version.

Original context

// GELU computes Gaussian Error Linear Unit
// GELU(x) = x · Φ(x) where Φ is the CDF of N(0,1)
func GELU(x float64) float64 {
	return 0.5 * x * (1.0 + math.Erf(x/math.Sqrt(2.0)))
}

// SwiGLU computes SwiGLU activation
// SwiGLU(x, W, V, b, c) = Swish(xW + b) ⊗ (xV + c)
func SwiGLU(x []float64, gate, linear []float64) []float64 {
	n := len(x)
	if n == 0 || len(gate) != n || len(linear) != n {
		return x
	}

	result := make([]float64, n)
	for i := range result {
		// Swish: x · sigmoid(x)
		swish := gate[i] * Sigmoid(gate[i])

SwiGLU(x, W, V, b, c) = Swish(xW + b) ⊗ (xV + c)

Related builds