Synthetic Pre-pretraining Survives Scale, but Not as a Grammatical Prior
Paper • 2609.39827 • Published • 12
Dataset Viewer
The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.
Synthetic Pre-pretraining Datasets
This repository contains pre-processed datasets from the paper Synthetic Pre-pretraining Survives Scale, but Not as a Grammatical Prior.
The datasets cover pre-pretraining (PPT) tasks such as k-Shuffle Dyck, Set, MP-Struct Core, and NCA, as well as control and pre-training (PT) mixtures (e.g., C4, SmolLM3, Olmo3, Marin, FineWeb-Edu). They are tokenized and used for training and evaluating language models across scales from 500M to 7B parameters.
- Paper: Hugging Face Paper | arXiv
- Code: GitHub repository
- Project page: verify-ppt
For preprocessing, training, evaluation, and the full list of dataset repositories, please refer to the GitHub README.
- Downloads last month
- 72