17. layers/softmax

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

When

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

Where

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

cd 17-softmax && source ../env.sh && go run .

Why

Classification heads and attention need stable softmax variants, including sparse/Gumbel/Entmax for research paths.

What

Weightless layer; KindStandard/Temperature/Grid/Hierarchical/Gumbel/Masked/Sparse/… WebGPU covers std family; exotic kinds hard-error on GPU (no silent host).

Sample output (captured)

[0.6380664 0.23473153 0.09543471 0.031767454] <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/17-softmax).

package main

import (
	"fmt"

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

func main() {
	l, err := softmax.New(softmax.Config{Dim: 4, Kind: softmax.KindStandard})
	if err != nil {
		panic(err)
	}
	x := core.NewTensor[float32](1, 4)
	copy(x.Data, []float32{2, 1, 0.1, -1})
	_, y, err := softmax.Forward(l, x)
	fmt.Println(y.Data, err)
}