Zequn Chen

Results 2 comments of Zequn Chen

Hi, don't know if you have solved the problem, I think it might be related to [this](https://github.com/huggingface/transformers/pull/10956) pull request. I solved the similar **NaN loss** problem with **t5** model when...

@pengzhangzhi Hi, thanks for your reply! I've just checked #2 and the Score SDE paper, and found your observation quite inspiring. It seems there's some inherent connection between probability flow...