Ouyang

icon company BiliBili icon location Shanghai

Results 1 comments of Ouyang

DKD: the nckd loss becomes infinity after one iteration of training.

After I adjusted the learning rate down, the loss is no longer nan, but the nckd loss is still 0. ![image](https://github.com/megvii-research/mdistiller/assets/67104283/dab3e482-cb4e-4108-a92d-38b91d807d69)