SWEBench-verified-mini is a subset of SWEBench-verified that uses 50 instead of 500 datapoints, requires 5GB instead of 130GB of storage and has approximately the same distribution of performance, test pass rates and difficulty as the original dataset.
You can find more details here:
https://github.com/mariushobbhahn/make_swe_bench_verified_mini
If you use the Inspect implementation, you can merely switch the dataset: str = "princeton-nlp/SWE-bench_Verified", to dataset: str =… See the full description on the dataset page:
https://huggingface.co/datasets/MariusHobbhahn/swe-bench-verified-mini.
