File size: 1,709 Bytes
aeab535
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
import torch
from safetensors.torch import load_file

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

def priority_encode(bits, weights):
    """8-to-3 priority encoder with valid bit.

    Returns (y2, y1, y0, valid) where y2y1y0 is the index of highest active input.

    Highest index (x7) has highest priority.

    """
    inp = torch.tensor([float(b) for b in bits])

    # Compute winner for each input position
    winners = []
    for i in range(8):
        win = int((inp * weights[f'winner{i}.weight']).sum() + weights[f'winner{i}.bias'] >= 0)
        winners.append(win)

    # Compute output bits from winners
    win_137 = torch.tensor([float(winners[i]) for i in [1,3,5,7]])
    y0 = int((win_137 * weights['y0.weight']).sum() + weights['y0.bias'] >= 0)

    win_2367 = torch.tensor([float(winners[i]) for i in [2,3,6,7]])
    y1 = int((win_2367 * weights['y1.weight']).sum() + weights['y1.bias'] >= 0)

    win_4567 = torch.tensor([float(winners[i]) for i in [4,5,6,7]])
    y2 = int((win_4567 * weights['y2.weight']).sum() + weights['y2.bias'] >= 0)

    valid = int((inp * weights['valid.weight']).sum() + weights['valid.bias'] >= 0)

    return y2, y1, y0, valid

if __name__ == '__main__':
    w = load_model()
    print('8-to-3 Priority Encoder (highest index wins)')
    print('Input -> Index, Valid')

    # Single-bit tests
    for i in range(8):
        bits = [0]*8
        bits[i] = 1
        y2, y1, y0, valid = priority_encode(bits, w)
        print(f'x{i} only: {y2*4 + y1*2 + y0}, {valid}')

    # No input
    y2, y1, y0, valid = priority_encode([0]*8, w)
    print(f'None: {y2*4 + y1*2 + y0}, valid={valid}')