Datasets:

Tasks:
Other
Modalities:
Text
Formats:
parquet
ArXiv:
Libraries:
Datasets
Dask
License:

Could this dataset be repurposed for LLM training?

#7
by cppowboy - opened

A benchmark size of 20k is a bit too large for evaluation.

This dataset is primarily designed for large-scale RL training.
For evaluation purposes, we maintain a separate, updatable benchmark (based on this data) here: nebius/SWE-rebench-leaderboard, which currently contains around 400 tasks.

Sign up or log in to comment