| import torch | |
| from safetensors.torch import load_file | |
| def load_model(path='model.safetensors'): | |
| return load_file(path) | |
| def compare4(a3, a2, a1, a0, b3, b2, b1, b0, weights): | |
| """4-bit magnitude comparator. Returns (GT, LT, EQ).""" | |
| inp = torch.tensor([float(a3), float(a2), float(a1), float(a0), | |
| float(b3), float(b2), float(b1), float(b0)]) | |
| gt = int((inp @ weights['gt.weight'].T + weights['gt.bias'] >= 0).item()) | |
| lt = int((inp @ weights['lt.weight'].T + weights['lt.bias'] >= 0).item()) | |
| gt_lt = torch.tensor([float(gt), float(lt)]) | |
| eq = int((gt_lt @ weights['eq.weight'].T + weights['eq.bias'] >= 0).item()) | |
| return gt, lt, eq | |
| if __name__ == '__main__': | |
| w = load_model() | |
| print('Comparator4bit examples:') | |
| for a, b in [(5, 3), (3, 5), (7, 7), (0, 15), (15, 0)]: | |
| a3, a2, a1, a0 = (a >> 3) & 1, (a >> 2) & 1, (a >> 1) & 1, a & 1 | |
| b3, b2, b1, b0 = (b >> 3) & 1, (b >> 2) & 1, (b >> 1) & 1, b & 1 | |
| gt, lt, eq = compare4(a3, a2, a1, a0, b3, b2, b1, b0, w) | |
| print(f' A={a:2d}, B={b:2d} -> GT={gt}, LT={lt}, EQ={eq}') | |