联想发布端侧AI新技术，宣布PC本地推理媲美 OpenAI-mini

Core Insights - Lenovo defines "super intelligent agents" with three core functions: perception and interaction, cognition and decision-making, and autonomy and evolution [1] - The company launched "Lenovo Tianxi" personal super intelligent agent and "Lenovo Lexiang" enterprise super intelligent agent at the 2025 Lenovo Tech World [1] Group 1 - Lenovo showcased its latest innovation in edge computing with the "Lenovo Inference Acceleration Engine," developed in collaboration with Tsinghua University and Wuwen Chip [3] - This inference engine is designed as a software-hardware collaborative optimization platform for efficient AI PC inference, enabling local inference capabilities on a standard PC to rival OpenAI's o1-mini cloud model [3] - The engine significantly enhances inference speed while reducing power consumption and memory usage through high-performance parallel decoding, operator fusion optimization, and heterogeneous computing technology [3][4] Group 2 - A demonstration at the event involved a "race" where three identical AI PCs solved a challenging math problem, showing that the combination of the second-generation model and the "Lenovo Inference Acceleration Engine" achieved a time of 13 seconds with a perfect score [3] - The advancements in model capabilities and the "Lenovo Inference Acceleration Engine" lead to faster pre-filling, quicker decoding speeds, and shorter inference times, all while consuming less memory and energy [4] - Lenovo's CEO, Yang Yuanqing, indicated that the next generation of AI PCs will soon showcase the powerful local inference capabilities of the "Lenovo Inference Acceleration Engine," which can match cloud-based large models at a lower cost [4]