DRIFT_QAFT is a Question-Answering dataset designed for the DRIFT project, which focuses on decoupling knowledge and reasoning in large language models.
This dataset is derived from the Wikipedia dataset released by Wikimedia on Hugging Face:
https://huggingface.co/datasets/wikimedia/wikipedia
The original data comes from Wikipedia snapshots provided by Wikimedia.
Entries are bucketed into long-context intervals based on the… See the full description on the dataset page:
https://huggingface.co/datasets/SII-LancelotXie/DRIFT_QAFT.
