title	Dual-Loop Cognitive Controller
emoji	🧠
colorFrom	indigo
colorTo	green
sdk	gradio
sdk_version	5.20.0
app_file	app.py
pinned	false
license	mit
short_description	Latent Deliberation & Memory Architecture for LLMs
models	Qwen/Qwen3.5-2B CH3NDev/dual-loop-qwen3.5-2b

Dual-Loop Cognitive Controller — Interactive Demo

This Hugging Face Space demonstrates the Dual-Loop Cognitive Controller on Qwen/Qwen3.5-2B ($D=2048$, Layer 11 hook).

Key Innovations Demonstrated:

1. Zero-Token Latent Deliberation (System 2): Recursively ponders in continuous latent space ($h \in \mathbb{R}^D$) inside GPU SRAM without emitting intermediate CoT tokens.
2. Cognitive Matrix Helper (Tversky Elimination-by-Aspects): Automatically prunes 40%–57% superficial distractor options (wrong logs) in Bench 1 and focuses System 2 deliberation in Bench 2, boosting accuracy from 50.0% to 83.3%.
3. Hippocampal Episodic Virtual Memory: Recalls previously verified reasoning traces in <0.01 seconds (3,146x speedup) with 0 FLOPs.

Visit the GitHub Repository or PyPI Package.