metadata
license: cc-by-4.0
task_categories:
- text-classification
language:
- en
configs:
- config_name: arxiv_mia
data_files: data/arxiv_mia.jsonl
default: true
- config_name: arxiv_mia_dev
data_files: data/arxiv_mia_dev.jsonl
- config_name: arxiv_mia_test
data_files: data/arxiv_mia_test.jsonl
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: Probing Language Models for Pre-training Data Detection