Hula0401/bench_score

该数据集包含用于BenchCAD/cad_bench_200基准测试的Per-(prediction-repo-sha × model) IoU-24分数。数据由`eval/remote_iou24_service.py`脚本更新，存储在cadrille仓库中。数据集分为详细数据（per_case文件夹中的文件，每个文件对应一个特定的sha和模型组合，包含每个stem的详细数据）和汇总数据（summary.parquet文件，按sha和模型分组汇总）。详细数据包括sha_short、模型、stem、code_hash、iou、iou_24、rot_idx、exec_ok、reason、latency_s、scored_at等字段；汇总数据包括sha_short、模型、n_total、n_exec_ok、exec_rate、mean_iou_24、median_iou_24、pass_iou_24_0.5、pass_iou_24_0.7、updated_at等字段。

This dataset contains Per-(prediction-repo-sha × model) IoU-24 scores for the BenchCAD/cad_bench_200 benchmark. The data is updated by the `eval/remote_iou24_service.py` script and stored in the cadrille repository. The dataset is divided into detailed data (files in the per_case folder, each corresponding to a specific sha and model combination, containing detailed data for each stem) and summary data (summary.parquet file, aggregated by sha and model). The detailed data includes fields such as sha_short, model, stem, code_hash, iou, iou_24, rot_idx, exec_ok, reason, latency_s, scored_at; the summary data includes fields such as sha_short, model, n_total, n_exec_ok, exec_rate, mean_iou_24, median_iou_24, pass_iou_24_0.5, pass_iou_24_0.7, updated_at.