Machine learning has achieved significant progress in recent years, with algorithms matching human capabilities in various tasks. However, the real challenge lies not just in training these models, but in utilizing them optimally in real-world applications. This is where inference in AI becomes crucial, emerging as a primary concern for scientists