13. layers/swiglu — gated FFN

Part: III · Layers
Package: github.com/openfluke/welvet/layers/swiglu
Status: ok — ✅

When

Building or training a net that needs the layers/swiglu Op (also usable as a Parallel cam).

Where

import "github.com/openfluke/welvet/layers/swiglu"

cd 13-swiglu && source ../env.sh && go run .

Why

Modern decoder FFNs are SiLU(gate)⊙up → down. Projections must share Dense’s quant/backend matrix.

What

Gate/Up/Down Dense children; DefaultFFN(dModel); WebGPU SiLU⊙ fuse on forward.

Sample output (captured)

64 <nil>

Live capture from go run ./cmd/runall on local ../../welvet (exit 0).

Source

Copied from the Welvet feature book examples (openfluke.github.io/welvet/examples/13-swiglu).

package main

import (
	"fmt"

	"github.com/openfluke/welvet/core"
	"github.com/openfluke/welvet/layers/swiglu"
)

func main() {
	l, err := swiglu.New(swiglu.DefaultFFN(64))
	if err != nil {
		panic(err)
	}
	x := core.NewTensor[float32](1, 64)
	_, post, err := swiglu.Forward(l, x)
	fmt.Println(len(post.Data), err)
}