What it is
A de-duplicated, provenance-tracked pool of 254,561 open-licensed passages labeled across 18 LLMs, filtered to match the Name That LLM competition setting. 254,561 unique passages from 59 open Kaggle datasets, labeled across 18 LLMs and 6 labs, with prompt groups, source provenance, train-similarity scores, and leakage/routing flags.
Key results
- 254,561 unique passages
- 18 model labels across 6 labs
- 59 source datasets, CC BY-SA 4.0
Useful for training and evaluating classifiers that identify which LLM generated a passage.