File size: 2,198 Bytes
96d9c09 | 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 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 | ---
license: mit
tags:
- pytorch
- safetensors
- threshold-logic
- neuromorphic
- comparison
---
# threshold-comparator8bit
8-bit magnitude comparator. Compares two 8-bit unsigned values.
## Function
compare8(A, B) -> (GT, LT, EQ)
- GT = 1 if A > B
- LT = 1 if A < B
- EQ = 1 if A = B
Exactly one output is always active.
## Architecture
```
A[7:0] B[7:0]
│ │
└───────┬───────┘
│
┌──────┴──────┐
│ │
▼ ▼
┌───┐ ┌───┐
│GT │ │LT │ Layer 1
│A-B│ │B-A│
│>=1│ │>=1│
└───┘ └───┘
│ │
└──────┬──────┘
│
▼
┌─────┐
│ EQ │ Layer 2
│ NOR │
└─────┘
```
Uses positional weighting: treats inputs as binary numbers.
**GT neuron:** weights A bits positively (128,64,32,16,8,4,2,1), B bits negatively.
Fires when weighted sum A - B >= 1.
**LT neuron:** opposite weights. Fires when B - A >= 1.
**EQ neuron:** NOR of GT and LT. Fires when both are zero.
## Parameters
| | |
|---|---|
| Inputs | 16 |
| Outputs | 3 |
| Neurons | 3 |
| Layers | 2 |
| Parameters | 37 |
| Magnitude | 1024 |
## Truth Table (examples)
| A | B | GT | LT | EQ |
|-----|-----|----|----|---|
| 0 | 0 | 0 | 0 | 1 |
| 100 | 50 | 1 | 0 | 0 |
| 50 | 100 | 0 | 1 | 0 |
| 255 | 255 | 0 | 0 | 1 |
## Usage
```python
from safetensors.torch import load_file
import torch
w = load_file('model.safetensors')
def compare(a, b):
a_bits = [(a >> (7-i)) & 1 for i in range(8)]
b_bits = [(b >> (7-i)) & 1 for i in range(8)]
inp = torch.tensor([float(x) for x in a_bits + b_bits])
gt = int((inp @ w['gt.weight'].T + w['gt.bias'] >= 0).item())
lt = int((inp @ w['lt.weight'].T + w['lt.bias'] >= 0).item())
eq = int((torch.tensor([float(gt), float(lt)]) @ w['eq.weight'].T + w['eq.bias'] >= 0).item())
return gt, lt, eq
# compare(200, 100) = (1, 0, 0) # 200 > 100
```
## License
MIT
|