create_safetensors.py

import torch
from safetensors.torch import save_file

# Priority encoder: outputs binary index of highest-set input
# Inputs: I3, I2, I1, I0 (I3 is highest priority)
# Outputs: Y1, Y0 (binary encoding)

weights = {}

# Y1: fires when I3 or I2 is set (highest bit is 2 or 3)
weights['y1.weight'] = torch.tensor([[1.0, 1.0, 0.0, 0.0]], dtype=torch.float32)
weights['y1.bias'] = torch.tensor([-1.0], dtype=torch.float32)

# Y0: fires when I3 is set, OR when I2 is NOT set but I1 is set
# This gives output bit 0 for indices 1 and 3
weights['y0.weight'] = torch.tensor([[3.0, -2.0, 1.0, 0.0]], dtype=torch.float32)
weights['y0.bias'] = torch.tensor([-1.0], dtype=torch.float32)

save_file(weights, 'model.safetensors')

def encode4to2(i3, i2, i1, i0):
    inp = torch.tensor([float(i3), float(i2), float(i1), float(i0)])
    y1 = int((inp @ weights['y1.weight'].T + weights['y1.bias'] >= 0).item())
    y0 = int((inp @ weights['y0.weight'].T + weights['y0.bias'] >= 0).item())
    return y1, y0

print("Verifying 4to2encoder...")
errors = 0
for val in range(16):
    i3, i2, i1, i0 = (val >> 3) & 1, (val >> 2) & 1, (val >> 1) & 1, val & 1
    y1, y0 = encode4to2(i3, i2, i1, i0)

    # Expected: binary of highest set bit position
    if i3: expected = (1, 1)      # 3
    elif i2: expected = (1, 0)    # 2
    elif i1: expected = (0, 1)    # 1
    else: expected = (0, 0)       # 0 or none

    if (y1, y0) != expected:
        errors += 1
        print(f"ERROR: I={i3}{i2}{i1}{i0} -> ({y1},{y0}), expected {expected}")

if errors == 0:
    print("All 16 test cases passed!")
print(f"Magnitude: {sum(t.abs().sum().item() for t in weights.values()):.0f}")