PrinceAnX/open_model_evolution_data

该数据集与论文《开放智能的经济学：追踪模型生态系统中的权力与参与度》一同发布，对开放模型经济中的集中动态和演变特征进行了严格审查。它汇编了从2025年2月至今的每周模型下载历史，以及来自Hugging Face模型中心的详细模型元数据。数据集涵盖851，000个模型，每个模型超过200个聚合属性，总计22亿次下载。数据来源于每周快照，并每日更新。文件all_downloads_with_annotations.parquet包含所有下载记录，one_year_rolling.parquet仅统计模型创建后一年内的下载次数。研究人员可利用此数据集分析：开放权重AI中经济权力的根本性再平衡；模型属性的统计显著变化，如平均模型大小的增加、多模态生成、量化和专家混合架构的增长；数据透明度趋势以及开发者中介的出现。项目还包括一个交互式仪表板，用于实时监控这些动态。This dataset, released in conjunction with the paper Economies of Open Intelligence: Tracing Power & Participation in the Model Ecosystem, provides a rigorous examination of concentration dynamics and evolving characteristics in the open model economy. It compiles a history of weekly model downloads (February 2025-Present) alongside detailed model metadata from the Hugging Face Model Hub. The dataset encompasses 851,000 models, over 200 aggregated attributes per model, and 2.2 billion downloads. The data is compiled from weekly snapshots and updated daily. `all_downloads_with_annotations.parquet` contains all downloads and `one_year_rolling.parquet` counts only downloads up to one year from model creation. Researchers can leverage this dataset to analyze: The fundamental rebalancing of economic power in open-weight AI; Statistically significant shifts in model properties, such as increases in average model size, growth in multimodal generation, quantization, and mixture-of-experts architectures; Trends in data transparency and the emergence of developer intermediaries. The project also features an interactive dashboard for real-time monitoring of these dynamics.