This dataset was generated from the public BABILong scripts and is intended for long-context QA1 workflows.
Each sample is a standalone JSON file for easy inspection.
Path pattern: samples/qa1/1000k/000000.json
No JSONL bundle required for browsing individual samples.
Task: qa1_single-supporting-fact
Context length: 1M
Sample count: 40
Noise corpus: wikitext-2 (used instead of PG19 for efficiency in this release)… See the full description on the dataset page:
https://huggingface.co/datasets/luisml77/babilong-qa1-1m.
