Will opensource pretrain or sft data?

[MonolithFoundation]

Will opensource pretrain or sft data?

[wenyihong]

Thanks for the question. The copyrights of training data belong to Zhipu AI. The training data (including pretrain and sft) construction process is similar to the original CogAgent paper, you can refer to https://arxiv.org/abs/2312.08914 for details. And we notice that many follow-ups use similar construction processes, and some of them haveopen-sourced the datasets.

[MonolithFoundation]

Can u suggestion some link for those opensourced data?