#Amax

'Amin is an optimistic, adaptive scheduler that uses lower-bound estimates and ordered eviction to cut LLM inference latency dramatically, often matching hindsight-optimal performance and outperforming conservative algorithms by up to 5x.'