计算机：蒸馏模型加速AI平权

Investment Rating - The report maintains an "Outperform" rating for the computer industry [2][10]. Core Insights - The report highlights that 2025 is expected to be a pivotal year for AI model accessibility, driven by advancements in model distillation technology, which significantly lowers the cost and barriers for AI model development [4][6]. - The research indicates that the AI inference model s1, trained at a cost of less than $50, demonstrates performance comparable to existing models, showcasing the effectiveness of the distillation process [4][5]. - The report emphasizes the potential for increased AI application and edge computing capabilities, as well as a revaluation of domestic computing power in light of the expected surge in inference computing [6]. Summary by Sections Market Performance - The computer sector has shown a performance range from -13% to +67% compared to the CSI 300 index over the specified periods [3]. Technological Developments - The report discusses the successful training of the AI inference model s1 using a small dataset and supervised fine-tuning, achieving a 27% performance improvement over the o1-preview model in competitive math problems [5]. Investment Recommendations - The report suggests focusing on companies involved in model distillation, AI applications, edge computing, and AI computing power, including firms like Zhixin Precision, Tuolisi, and Kingsoft Office among others [6].