threshold-minority / model.py
phanerozoic's picture
Rename from tiny-Minority-verified
b510417 verified
"""
Threshold Network for Minority Gate
A formally verified single-neuron threshold network computing 8-bit minority.
Outputs 1 when 3 or fewer of the 8 inputs are true (strict minority).
"""
import torch
from safetensors.torch import load_file
class ThresholdMinority:
"""
Minority gate implemented as a threshold neuron.
Circuit: output = (sum of weighted inputs + bias >= 0)
With all weights=-1, bias=3: fires when hamming weight <= 3.
"""
def __init__(self, weights_dict):
self.weight = weights_dict['weight']
self.bias = weights_dict['bias']
def __call__(self, bits):
inputs = torch.tensor([float(b) for b in bits])
weighted_sum = (inputs * self.weight).sum() + self.bias
return (weighted_sum >= 0).float()
@classmethod
def from_safetensors(cls, path="model.safetensors"):
return cls(load_file(path))
def forward(x, weights):
"""
Forward pass with Heaviside activation.
Args:
x: Input tensor of shape [..., 8]
weights: Dict with 'weight' and 'bias' tensors
Returns:
1 if minority (3 or fewer of 8) are true, else 0
"""
x = torch.as_tensor(x, dtype=torch.float32)
weighted_sum = (x * weights['weight']).sum(dim=-1) + weights['bias']
return (weighted_sum >= 0).float()
if __name__ == "__main__":
weights = load_file("model.safetensors")
model = ThresholdMinority(weights)
print("Minority Gate Tests:")
print("-" * 40)
test_cases = [
[0,0,0,0,0,0,0,0],
[1,0,0,0,0,0,0,0],
[1,1,1,0,0,0,0,0],
[1,1,1,1,0,0,0,0],
[1,1,1,1,1,0,0,0],
[1,1,1,1,1,1,1,1],
]
for bits in test_cases:
hw = sum(bits)
out = int(model(bits).item())
expected = 1 if hw <= 3 else 0
status = "OK" if out == expected else "FAIL"
print(f"HW={hw}: Minority({bits}) = {out} [{status}]")