Upload folder using huggingface_hub

README.md

@@ -10,25 +10,25 @@ tags:
 
 # threshold-lessthan
 
-4-bit less-than comparator. Returns 1 if a < b, 0 otherwise.
+8-bit less-than comparator. Returns 1 if a < b, 0 otherwise.
 
 ## Circuit
 
 ```
-  a0 b0 0         a1 b1          a2 b2          a3 b3
-   │  │ │          │  │           │  │           │  │
-   └──┼─┘          └──┼──┘        └──┼──┘        └──┼──┘
-      ▼               ▼              ▼              ▼
-  ┌───────┐       ┌───────┐      ┌───────┐      ┌───────┐
-  │  FS0  │──b0───│  FS1  │──b1──│  FS2  │──b2──│  FS3  │───► (a < b)
-  └───────┘       └───────┘      └───────┘      └───────┘
-      │               │              │              │
-      ▼               ▼              ▼              ▼
-     d0              d1             d2             d3
-   (unused)       (unused)       (unused)       (unused)
+a0 b0    a1 b1    a2 b2    a3 b3    a4 b4    a5 b5    a6 b6    a7 b7
+ │ │ 0    │ │      │ │      │ │      │ │      │ │      │ │      │ │
+ └┬┘│     └┬┘      └┬┘      └┬┘      └┬┘      └┬┘      └┬┘      └┬┘
+  ▼ │      ▼        ▼        ▼        ▼        ▼        ▼        ▼
+┌───┐│  ┌───┐    ┌───┐    ┌───┐    ┌───┐    ┌───┐    ┌───┐    ┌───┐
+│FS0│┴──│FS1│────│FS2│────│FS3│────│FS4│────│FS5│────│FS6│────│FS7│──►(a<b)
+└───┘   └───┘    └───┘    └───┘    └───┘    └───┘    └───┘    └───┘
+  │       │        │        │        │        │        │        │
+  ▼       ▼        ▼        ▼        ▼        ▼        ▼        ▼
+ d0      d1       d2       d3       d4       d5       d6       d7
+      (difference bits unused - only final borrow matters)
 ```
 
-Uses 4 cascaded full subtractors. The final borrow output indicates a < b.
+Uses 8 cascaded full subtractors. The final borrow output indicates a < b.
 
 ## Mechanism
 
@@ -37,7 +37,7 @@ The circuit computes a - b using subtraction:
 - If a ≥ b: no borrow propagates, result is non-negative
 - If a < b: borrow propagates out, indicating underflow
 
-The difference bits (d0-d3) are computed but unused. Only the final borrow matters.
+The difference bits (d0-d7) are computed but unused. Only the final borrow matters.
 
 ## Truth Table (sample)
 
@@ -46,13 +46,11 @@ The difference bits (d0-d3) are computed but unused. Only the final borrow matte
 | 0 | 0 | 0 |
 | 0 | 1 | 1 |
 | 1 | 0 | 0 |
-| 5 | 10 | 1 |
-| 10 | 5 | 0 |
-| 7 | 7 | 0 |
-| 15 | 0 | 0 |
-| 0 | 15 | 1 |
-
-All 256 input combinations verified correct.
+| 127 | 128 | 1 |
+| 255 | 0 | 0 |
+| 0 | 255 | 1 |
+| 100 | 100 | 0 |
+| 99 | 100 | 1 |
 
 ## Full Subtractor Structure
 
@@ -88,12 +86,12 @@ All 256 input combinations verified correct.
 
 ## Architecture
 
-| Component | Per FS | Total (4 FSs) |
+| Component | Per FS | Total (8 FSs) |
 |-----------|--------|---------------|
-| Neurons | 9 | 36 |
-| Parameters | 27 | 108 |
+| Neurons | 9 | 72 |
+| Parameters | 27 | 216 |
 
-**Layers: 16** (4 FSs × 4 layers each)
+**Layers: 32** (8 FSs × 4 layers each)
 
 ## Comparison Family
 
@@ -101,7 +99,7 @@ All 256 input combinations verified correct.
 |---------|-----------|----------------|
 | **LessThan** | a < b | borrow_out of (a - b) |
 | Equal | a = b | NOR of all XOR bits |
-| GreaterThan | a > b | NOT(a < b) AND NOT(a = b) |
+| GreaterThan | a > b | LessThan(b, a) |
 
 ## Usage
 
@@ -111,14 +109,14 @@ from safetensors.torch import load_file
 w = load_file('model.safetensors')
 
 def less_than(a, b):
-    """a, b: 4-bit lists [a0,a1,a2,a3] (LSB first)
+    """a, b: 8-bit lists [a0..a7] (LSB first)
     Returns: 1 if a < b, 0 otherwise"""
     # See model.py for full implementation
     pass
 
-# Example: 5 < 10?
-a = [1, 0, 1, 0]  # 5 in LSB-first
-b = [0, 1, 0, 1]  # 10 in LSB-first
+# Example: 99 < 100?
+a = [(99 >> i) & 1 for i in range(8)]
+b = [(100 >> i) & 1 for i in range(8)]
 result = less_than(a, b)  # Returns 1
 ```
 

config.json

@@ -1,9 +1,9 @@
 {
   "name": "threshold-lessthan",
-  "description": "4-bit less-than comparator as threshold circuit",
-  "inputs": 8,
+  "description": "8-bit less-than comparator as threshold circuit",
+  "inputs": 16,
   "outputs": 1,
-  "neurons": 36,
-  "layers": 16,
-  "parameters": 108
+  "neurons": 72,
+  "layers": 32,
+  "parameters": 216
 }

model.py

@@ -18,23 +18,23 @@ def full_sub(a, b, bin_in, w, prefix):
     return int(d), bout
 
 def less_than(a, b, weights):
-    """4-bit less-than comparator.
-    a, b: lists of 4 bits each (LSB first)
+    """8-bit less-than comparator.
+    a, b: lists of 8 bits each (LSB first)
     Returns: 1 if a < b, 0 otherwise
     """
     borrows = [0]
-    for i in range(4):
+    for i in range(8):
         d, bout = full_sub(a[i], b[i], borrows[i], weights, f'fs{i}')
         borrows.append(bout)
-    return borrows[4]
+    return borrows[8]
 
 if __name__ == '__main__':
     w = load_model()
-    print('4-bit LessThan Comparator')
+    print('8-bit LessThan Comparator')
     print('a < b tests:')
-    tests = [(0, 0), (0, 1), (1, 0), (5, 10), (10, 5), (7, 7), (15, 0), (0, 15)]
+    tests = [(0, 0), (0, 1), (1, 0), (127, 128), (255, 0), (0, 255), (100, 100), (99, 100)]
     for a_val, b_val in tests:
-        a = [(a_val >> i) & 1 for i in range(4)]
-        b = [(b_val >> i) & 1 for i in range(4)]
+        a = [(a_val >> i) & 1 for i in range(8)]
+        b = [(b_val >> i) & 1 for i in range(8)]
         result = less_than(a, b, w)
-        print(f'{a_val:2d} < {b_val:2d} = {result}')
+        print(f'{a_val:3d} < {b_val:3d} = {result}')

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9221161e5cbde88f0223631b5ef8100ea0cea9cc8719fd809191181c5315cb27
-size 4808
+oid sha256:bfb1c6effebca579db177baa9ca4590908758f0dd15dba0e3c83d01c8e00972a
+size 9632