[long-context models trained with "original text paraphrasing" dataset](https://github.com/yuyijiong/train_with_paraphrasing)