HumanEval-Sci: a verification-grounded benchmark for scientific code generation
收藏官方服务：
资源简介：
A benchmark of 73 scientific-code prompts across seven domains (physics, chemistry, biology, climate, mathematics, numerical methods, and engineering). Each prompt carries a reference solution, executable tests, and per-prompt verification targets (dimensional vectors, declared limits, and validation envelopes). The repository also includes the recorded evaluation runs behind the HumanEval-Sci pilot study and a dependency-free script that regenerates the study's supplementary tables from those records.
提供机构：
Zenodo创建时间：
2026-05-27
