model.py

import torch
from safetensors.torch import load_file

def load_model(path='model.safetensors'):
    return load_file(path)

def canalizing(x2, x1, x0, weights):
    """Canalizing function: if x0=1, output=0 regardless of x1,x2."""
    inp = torch.tensor([float(x2), float(x1), float(x0)])

    not_x0 = int((inp @ weights['not_x0.weight'].T + weights['not_x0.bias'] >= 0).item())
    or_x1_x2 = int((inp @ weights['or_x1_x2.weight'].T + weights['or_x1_x2.bias'] >= 0).item())

    l1 = torch.tensor([float(not_x0), float(or_x1_x2)])
    y = int((l1 @ weights['y.weight'].T + weights['y.bias'] >= 0).item())

    return y

if __name__ == '__main__':
    w = load_model()
    print('Canalizing Function (x0 canalizes to 0):')
    for i in range(8):
        x2, x1, x0 = (i >> 2) & 1, (i >> 1) & 1, i & 1
        y = canalizing(x2, x1, x0, w)
        print(f'  {x2}{x1}{x0} -> {y}')