You can deploy this model and start earning money today!

phanerozoic/
threshold-demux16

Name: threshold-demux16
Availability: InStock

Views

No views yet

threshold-demux16

1:16 demultiplexer. Routes data input to one of 16 outputs based on 4-bit select.

Function

DEMUX16(d, s3, s2, s1, s0) -> [y0..y15]

yi = d AND (s == i), where s = 8s3 + 4s2 + 2*s1 + s0

Architecture

Single layer with 16 neurons. Each output yi fires when d=1 AND select matches i.

Each neuron has:

Weight on d: +1
Weights on select bits: +1 if that bit is 1 in i, else -1
Bias: -(1 + popcount(i))

Parameters


Inputs	5 (1 data + 4 select)
Outputs	16
Neurons	16
Layers	1
Parameters	96
Magnitude	128

Usage

python

from safetensors.torch import load_file
import torch

w = load_file('model.safetensors')

def demux16(d, s3, s2, s1, s0):
    inp = torch.tensor([float(d), float(s3), float(s2), float(s1), float(s0)])
    return [int((inp * w[f'y{i}.weight']).sum() + w[f'y{i}.bias'] >= 0)
            for i in range(16)]

# Route d=1 to output 10 (s=1010)
print(demux16(1, 1, 0, 1, 0))  # [0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0]

License

MIT

phanerozoic/threshold-demux16