Skip to content

Dataset · shhu2001

Verified Scientific Code Generation Benchmark

shhu2001/SciCode-Verified

A manually corrected and verified version of the SciCode benchmark across 64 scientific programming tasks.

What's inside

An English text dataset containing 64 main tasks and 287 evaluated subtasks (under 1000 total records). The authors performed a line-by-line audit of the original SciCode test, found 263 defects, and fixed all confirmed errors while excluding one incorrect task.

What you can do with it

  • Fair evaluation of LLMs on scientific code writing tasks
  • Debugging and validating program code generation systems
Tasks
Text generation
Languages
English
Size
up to 1K rows
Downloads this month
44.1K
Likes
2

How to load it

from datasets import load_dataset

ds = load_dataset("shhu2001/SciCode-Verified")