import torch
from safetensors.torch import load_file

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

def clz4(a3, a2, a1, a0, w):
    """Count leading zeros. Returns 3-bit binary count (0-4)."""
    inp = torch.tensor([float(a3), float(a2), float(a1), float(a0)])

    # Layer 1
    has3 = int((inp @ w['has3.weight'].T + w['has3.bias'] >= 0).item())
    has2_first = int((inp @ w['has2_first.weight'].T + w['has2_first.bias'] >= 0).item())
    has1_first = int((inp @ w['has1_first.weight'].T + w['has1_first.bias'] >= 0).item())
    has0_first = int((inp @ w['has0_first.weight'].T + w['has0_first.bias'] >= 0).item())
    all_zero = int((inp @ w['all_zero.weight'].T + w['all_zero.bias'] >= 0).item())

    # Layer 2
    l1 = torch.tensor([float(has3), float(has2_first), float(has1_first), float(has0_first), float(all_zero)])
    y0 = int((l1 @ w['y0.weight'].T + w['y0.bias'] >= 0).item())
    y1 = int((l1 @ w['y1.weight'].T + w['y1.bias'] >= 0).item())
    y2 = int((l1 @ w['y2.weight'].T + w['y2.bias'] >= 0).item())

    return [y2, y1, y0]

if __name__ == '__main__':
    w = load_model()
    print('clz4 truth table:')
    for i in range(16):
        a3, a2, a1, a0 = (i >> 3) & 1, (i >> 2) & 1, (i >> 1) & 1, i & 1
        result = clz4(a3, a2, a1, a0, w)
        clz_val = result[0] * 4 + result[1] * 2 + result[2]
        print(f'  {a3}{a2}{a1}{a0} -> clz={clz_val}')