Phi-4-Mini

A 5-condition ablation study (3,175 total runs) validates that growth-ratio normalized Lyapunov energy functions serve as precise leading indicators of agent task failure, achieving zero stability vio…

Small language models (SLMs) are increasingly powering next-generation AI agents, challenging the assumption that larger models are always better. NVIDIA research shows SLMs excel at repetitive, speci…