Catmanjoe/mca-bdh-continual

MCA-BDH Continual Learning Model

Model Description

Brain-inspired Dynamic Hebbian (BDH) model with continual learning capabilities.

Architecture

• Neurons: 2048
• Layers: 8
• Low-rank dimension: 512
• Concepts: 256
• Slots: 16

Training

• Dataset: wikitext-2
• Checkpoint: wikitext-2_epoch1
• Timestamp: 2025-10-31T17:28:57.477209

Files

• checkpoints/*.pt - Full checkpoints with optimizer states (for resume training)
• model/*_model.pt - Model weights only (for inference)
• tracking/rules_*.db - Extracted symbolic rules database

Usage

import torch
from bdh.bdh_model import BDHModel
from bdh.symbolic_layer import SymbolicLayer

# Load model weights
checkpoint = torch.load('model/wikitext-2_epoch1_model.pt')
bdh_model = BDHModel(...)  # Initialize with config
bdh_model.load_state_dict(checkpoint['bdh_model'])
symbolic_layer = SymbolicLayer(...)
symbolic_layer.load_state_dict(checkpoint['symbolic_layer'])