遇见数据集

1337xyz1337xyz/sciknoweval-v2-hard-autogradable-512-2026-04-28

收藏
Hugging Face2026-04-28 更新2026-05-03 收录
官方服务:

资源简介:

SciKnowEval v2 Hard Autogradable 512 - 2026-04-28是一个包含512个例子的子集,来源于`hicai-zju/SciKnowEval`的`v2`版本中的`test`部分,用于Plan-CRL的科学推理评估。数据集通过特定的种子(20260428)选择样本,并进行了过滤和平衡处理,排除了`L1`级别,保留了`L2`、`L3`、`L4`级别,以及特定的自动评分类型(如多选题、判断题等)。数据集在四个科学领域(生物学、化学、材料、物理)中每个领域平衡包含128个例子,每个领域内进一步细分了不同难度级别的数量(32个L2、48个L3、48个L4)。此外,数据集提供了用于手动检查的字段和额外文件,如元数据摘要和选择清单。

SciKnowEval v2 Hard Autogradable 512 - 2026-04-28 is a 512-example sanity subset sampled from `hicai-zju/SciKnowEval` (`v2`, `test`) for Plan-CRL scientific reasoning evals. The subset was selected using a specific seed (20260428) and underwent filtering and balancing to exclude `L1`, retain `L2`, `L3`, `L4`, and specific autogradable types (e.g., multiple-choice questions, true/false questions). The dataset is balanced across four domains (Biology, Chemistry, Material, Physics) with 128 examples each, further divided by difficulty levels (32 L2, 48 L3, 48 L4 per domain). It also includes useful fields for manual inspection and extra files like metadata summaries and selection manifests.

提供机构:
1337xyz1337xyz
二维码
社区交流群
二维码
科研交流群
商业服务