Prompt caching을 통한 LLM 추론 비용 36% 절감

Cutting LLM inference costs by 36% with prompt caching