Efficient LLM Reasoning

Interpretable adaptive sampling for LLM test-time scaling and test-time compute. A fuzzy controller uses prompt complexity and model confidence to dynamically allocate inference-time compute for efficient LLM reasoning.