Dataset · nebius
Python GitHub Repository Issue and Pull Request Pairs
nebius/SWE-bench-extra
6,415 GitHub tasks and their solutions for training and evaluating software engineering AI agents.
What's inside
The text dataset contains 6,415 pairs of problem descriptions (Issues) and their corresponding Pull Requests. The data was collected from 1,988 open Python repositories on GitHub following the methodology of the well-known SWE-bench benchmark. The card also indicates that a larger and improved version of this set, SWE-rebench, has already been released.
What you can do with it
- Training and fine-tuning (SFT) autonomous coding AI agents
- Evaluating the ability of language models to fix bugs in real repositories
- Size
- 1K–10K rows
- Downloads this month
- 15.6K
- Likes
- 48
How to load it
from datasets import load_dataset
ds = load_dataset("nebius/SWE-bench-extra")