metadata
license: cc-by-4.0
task_categories:
- text-classification
language:
- en
Dataset Card for ArxivMIA
To evaluate various pre-training data detection methods in a more challenging scenario, we introduce ArxivMIA, a new benchmark comprising abstracts from the fields of Computer Science (CS) and Mathematics (Math) sourced from Arxiv.
- Repository: https://github.com/zhliu0106/probing-lm-data
- Paper [optional]: Probing Language Models for Pre-training Data Detection