metadata
license: other
library_name: transformers
tags:
- reasoning
- context-learning
- pretraining
- synthetic-data
- transformers
0.9zoo_op2-20+0.1teacher_op2
90% zoo op2-20, 10% teacher op2 pretraining mixture. This directory contains the final op2-only pretraining checkpoint and corresponding final RL checkpoints.
This is a context B pretraining checkpoint where the teacher component uses only op2.
Citation
@misc{zhang2025interplaypretrainingmidtrainingrl,
title={On the Interplay of Pre-Training, Mid-Training, and RL on Reasoning Language Models},
author={Charlie Zhang and Graham Neubig and Xiang Yue},
year={2025},
eprint={2512.07783},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2512.07783},
}