Oluben/COPAL-Continual-LLM-Adaptation
0
COPAL — Continual Off-Policy Adaptation for LLMs
Overview
Five-component continual RL framework for LLM adaptation.
Key Results
- 0.00% catastrophic forgetting
- FWT = +0.050 positive forward transfer
- 4.24x sample efficiency over PPO
- 77.7% hallucination reduction
Status
- ARC Prize 2026: Active submission
- arXiv paper: Coming soon