4-bit bit reversal, magnitude 8

README.md

@@ -0,0 +1,65 @@
+---
+license: mit
+tags:
+- pytorch
+- safetensors
+- threshold-logic
+- neuromorphic
+---
+
+# threshold-reverse4
+
+4-bit bit reversal. Reverses the order of bits.
+
+## Function
+
+reverse4(a3, a2, a1, a0) = [a0, a1, a2, a3]
+
+| Input | Output |
+|-------|--------|
+| 0001 | 1000 |
+| 1000 | 0001 |
+| 0110 | 0110 |
+| 1010 | 0101 |
+
+## Architecture
+
+Single layer with 4 neurons, each copying one input bit to its reversed position.
+
+| Output | Copies from | Weights [a3,a2,a1,a0] | Bias |
+|--------|-------------|------------------------|------|
+| y3 | a0 | [0, 0, 0, 1] | -1 |
+| y2 | a1 | [0, 0, 1, 0] | -1 |
+| y1 | a2 | [0, 1, 0, 0] | -1 |
+| y0 | a3 | [1, 0, 0, 0] | -1 |
+
+## Parameters
+
+| | |
+|---|---|
+| Inputs | 4 |
+| Outputs | 4 |
+| Neurons | 4 |
+| Layers | 1 |
+| Parameters | 8 |
+| Magnitude | 8 |
+
+## Usage
+
+```python
+from safetensors.torch import load_file
+import torch
+
+w = load_file('model.safetensors')
+
+def reverse4(a3, a2, a1, a0):
+    inp = torch.tensor([float(a3), float(a2), float(a1), float(a0)])
+    return [int((inp @ w[f'y{i}.weight'].T + w[f'y{i}.bias'] >= 0).item())
+            for i in [3, 2, 1, 0]]
+
+print(reverse4(1, 0, 0, 0))  # [0, 0, 0, 1]
+```
+
+## License
+
+MIT

config.json

@@ -0,0 +1,9 @@
+{
+  "name": "threshold-reverse4",
+  "description": "4-bit bit reversal",
+  "inputs": 4,
+  "outputs": 4,
+  "neurons": 4,
+  "layers": 1,
+  "parameters": 8
+}

create_safetensors.py

@@ -0,0 +1,53 @@
+import torch
+from safetensors.torch import save_file
+
+weights = {}
+
+# Input order: a3, a2, a1, a0
+# Output: y3=a0, y2=a1, y1=a2, y0=a3
+# Each output copies from reversed position
+
+# y3 = a0 (copy from position 3)
+weights['y3.weight'] = torch.tensor([[0.0, 0.0, 0.0, 1.0]], dtype=torch.float32)
+weights['y3.bias'] = torch.tensor([-1.0], dtype=torch.float32)
+
+# y2 = a1 (copy from position 2)
+weights['y2.weight'] = torch.tensor([[0.0, 0.0, 1.0, 0.0]], dtype=torch.float32)
+weights['y2.bias'] = torch.tensor([-1.0], dtype=torch.float32)
+
+# y1 = a2 (copy from position 1)
+weights['y1.weight'] = torch.tensor([[0.0, 1.0, 0.0, 0.0]], dtype=torch.float32)
+weights['y1.bias'] = torch.tensor([-1.0], dtype=torch.float32)
+
+# y0 = a3 (copy from position 0)
+weights['y0.weight'] = torch.tensor([[1.0, 0.0, 0.0, 0.0]], dtype=torch.float32)
+weights['y0.bias'] = torch.tensor([-1.0], dtype=torch.float32)
+
+save_file(weights, 'model.safetensors')
+
+# Verify
+def reverse4(a3, a2, a1, a0):
+    inp = torch.tensor([float(a3), float(a2), float(a1), float(a0)])
+    y3 = int((inp @ weights['y3.weight'].T + weights['y3.bias'] >= 0).item())
+    y2 = int((inp @ weights['y2.weight'].T + weights['y2.bias'] >= 0).item())
+    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 [y3, y2, y1, y0]
+
+print("Verifying reverse4...")
+errors = 0
+for i in range(16):
+    a3, a2, a1, a0 = (i >> 3) & 1, (i >> 2) & 1, (i >> 1) & 1, i & 1
+    result = reverse4(a3, a2, a1, a0)
+    expected = [a0, a1, a2, a3]
+    if result != expected:
+        errors += 1
+        print(f"ERROR: {a3}{a2}{a1}{a0} -> {result}, expected {expected}")
+
+if errors == 0:
+    print("All 16 test cases passed!")
+else:
+    print(f"FAILED: {errors} errors")
+
+mag = sum(t.abs().sum().item() for t in weights.values())
+print(f"Magnitude: {mag:.0f}")

model.py

@@ -0,0 +1,22 @@
+import torch
+from safetensors.torch import load_file
+
+def load_model(path='model.safetensors'):
+    return load_file(path)
+
+def reverse4(a3, a2, a1, a0, w):
+    """Reverse bit order of 4-bit input."""
+    inp = torch.tensor([float(a3), float(a2), float(a1), float(a0)])
+    y3 = int((inp @ w['y3.weight'].T + w['y3.bias'] >= 0).item())
+    y2 = int((inp @ w['y2.weight'].T + w['y2.bias'] >= 0).item())
+    y1 = int((inp @ w['y1.weight'].T + w['y1.bias'] >= 0).item())
+    y0 = int((inp @ w['y0.weight'].T + w['y0.bias'] >= 0).item())
+    return [y3, y2, y1, y0]
+
+if __name__ == '__main__':
+    w = load_model()
+    print('reverse4 examples:')
+    for val in [0b0001, 0b1000, 0b0110, 0b1010, 0b1111]:
+        a3, a2, a1, a0 = (val >> 3) & 1, (val >> 2) & 1, (val >> 1) & 1, val & 1
+        result = reverse4(a3, a2, a1, a0, w)
+        print(f'  {a3}{a2}{a1}{a0} -> {"".join(map(str, result))}')