CPT 训练数据 SFT-80K baseline
收藏官方服务：
资源简介：
# cpt-trainingdata-sft80k The 80K answerable-math SFT corpus used to train the `SFT-80K+RL` baseline in [Cognitive Pairwise Training (CPT)](https://github.com/Tsinghua-dhy/CPT) (paper §4.1). - 79,836 traces, volume-matched to the total token budget of `CPT + Math-SFT`. Train script: [`train/sft/run_sft80k_qwen3_4b.sh`](https://github.com/Tsinghua-dhy/CPT). ## Files - `train.parquet` — ~828 MB
提供机构：
maas创建时间：
2026-06-01
