laion/

nemotron-terminal-data_science

Per-source partition of nvidia/Nemotron-Terminal-Corpus, filtered to source == "data_science". The difficulty column preserves the original easy / medium / mixed split (na for the dataset_adapters/* files, which did not carry a difficulty label). Partitioning scheme:

adapters_{code,math,swe} — rows from dataset_adapters/{code,math,swe}.parquet {skill} (e.g. debugging, security, …) — rows from synthetic_tasks/skill_based/{easy,medium… See the full description on the dataset page: https://huggingface.co/datasets/laion/nemotron-terminal-data_science.