| # LightTiny 4M - Level 5 Field Dynamics Model |
|
|
| ## Overview |
|
|
| LightTiny is a proof-of-concept implementation of the LIGHTBRAIN architecture |
| using Level 5 Pure Field Dynamics. |
|
|
| ## Architecture |
|
|
| **5-Layer Compression:** |
| 1. **Field Configuration Storage** - Substrate field, attractors, coupling maps |
| 2. **Pattern Propagation Router** - Token embeddings, field projections |
| 3. **Resonance Tiers** - Sparse activation management |
| 4. **Field Dynamics Engine** - Field evolution parameters |
| 5. **Bridge + API** - Pattern to token mapping |
|
|
| ## Model Statistics |
|
|
| - Total Configurations: 3,997,696 |
| - Total Parameters: 7,102,464 |
| - Memory: 27.09 MB |
| - Vocab Size: 8000 |
| - Field Size: 256 x 488 |
| - Regions: 32 |
|
|
| ## Key Features |
|
|
| - **Sparse Activation**: Only 5.0% of field active at any time |
| - **No Backprop Training**: Uses resonance alignment (Hebbian-like learning) |
| - **Pure Field Dynamics**: Parameters are field configurations, not weights |
|
|
| ## Usage |
|
|
| ```python |
| from lightTiny.model.field_model import LightTinyModel |
| |
| model = LightTinyModel.load("exports/lighttiny-4m") |
| result = model.generate_token([2], temperature=0.8) |
| print(result.token_id) |
| ``` |
|
|
| ## License |
|
|
| Research use only. |
|
|
