Datasets:
metadata
language:
- en
size_categories:
- 10K<n<100K
license: apache-2.0
This is the RL training data for repro-rephraser-4B from RePro: Training Language Models to Faithfully Recycle the Web for Pretraining.
Each entry contains two columns, dataman_score and text.