Federated Reasoning Distillation Framework with Model Learnability-Aware Data Allocation
• Addresses bidirectional model learnability gap in federated LLM-SLM reasoning collaboration. • Introduces LaDa framework with learnability-aware data filter for high-reward sampl