hf-hub-daily-downloads-full
收藏资源简介:
Hugging Face Hub每日下载数据集记录了整个Hugging Face Hub上模型和数据集仓库的每日下载次数。该数据集通过比较每日元数据快照中累计的下载总量计数器,重建了每日下载序列,弥补了Hub API仅提供累计总量和滚动30天总量的限制。数据集包含两个主要文件:models.parquet覆盖模型仓库,包含约7620万行数据,涉及303万个仓库,数据起始于2025年2月28日;datasets.parquet覆盖数据集仓库,包含约6960万行数据,涉及124万个仓库,数据起始于2025年3月3日。两个文件的数据均持续更新至2026年7月15日(约400个观察日)。数据模式包含8个字段:repo_id(仓库名称,可能因重命名而变化)、hub_id(跨重命名稳定的唯一标识符)、date(UTC日期)、downloads_delta(自上次观察以来的下载增量)、span_days(距上次观察的天数)、is_first_seen(是否首次出现)、downloads_total(截至该日的累计下载总量)、downloads_30d(Hub的滚动30天计数器)和created_at(仓库创建日期)。数据行表示仓库的累计计数器发生了移动:在截至date的span_days天内发生了downloads_delta次下载。没有某日的数据行意味着该日下载量为零。该数据集适用于下载趋势分析、仓库流行度排名、时间序列建模等任务,并附带按总下载量降序排序的名称索引文件,便于快速检索和自动补全。
The Hugging Face Hub daily downloads dataset records the daily download counts for model and dataset repositories across the entire Hugging Face Hub. It reconstructs daily download sequences by comparing the cumulative total downloads counter in daily metadata snapshots, addressing the limitation of the Hub API, which only provides cumulative totals and rolling 30-day totals. The dataset includes two main files: models.parquet covers model repositories, containing approximately 76.2 million rows across 3.03 million repositories, with data starting from February 28, 2025; datasets.parquet covers dataset repositories, containing approximately 69.6 million rows across 1.24 million repositories, with data starting from March 3, 2025. Both files are continuously updated until July 15, 2026 (approximately 400 observation days). The data schema comprises 8 fields: repo_id (repository name, which may change due to renaming), hub_id (a unique identifier stable across renames), date (UTC date), downloads_delta (download increment since the last observation), span_days (days since the last observation), is_first_seen (whether it is the first appearance), downloads_total (cumulative total downloads up to that date), downloads_30d (Hubs rolling 30-day counter), and created_at (repository creation date). Each data row indicates a movement in the cumulative counter for a repository: downloads_delta downloads occurred within span_days up to date. The absence of a row for a specific date means zero downloads on that day. This dataset is suitable for download trend analysis, repository popularity ranking, time series modeling, and other tasks, and includes a name index file sorted by total downloads in descending order for quick retrieval and autocompletion.





