Jian Guan
What do you mean by "some codes must be modified"?
@BinbinBian I'preparing the code. I will send the code to you as soon as possible.
@qywu I guess it maybe result from the issue of dataset splits. ROCStories corpus contains 98,162 short stories as training data, and an additional 1,817 stories for validation and test,...
@TobiasLee I mean that we have mistaken the bleu score for modified n-gram precision score in the official publication. But we have update it in arxiv preprint version. Thank you!
@TobiasLee Yes.
@vinayshekharcmu We implement bleu based on the [open source code]( https://github.com/vikasnar/Bleu). And I am reruning our code and metrices in nltk to make sure we do this right. Thank you!
Thank you for your attention. You can find the commonsense data [here](https://cloud.tsinghua.edu.cn/d/670f7787b6554f308226/).
Thank you for your question! Although the key motivation of the reconstruction task is to provide more specific supervision signals for recognizing errors, UNION can generate meaningful editing results from...
感谢关注！填空任务和续写任务是LongLM的预训练任务，可以直接不经过finetune就进行生成。OutGen任务是我们设计LOT benchmark中的的一个下游任务，需要对LongLM在相应的训练集上进行训练才能正确生成。
@ouhongxu 请参考[示例生成代码](https://github.com/thu-coai/LOT-LongLM/blob/main/baselines/generation/gen.py)