Bocchi the Rock comments

Repositories
Issues
Comments

Results 2 comments of


                                            Bocchi the Rock

多进程跑出现loss为0，eval loss为nan

我也是遇到了训练过程中的loss 和 learning rate始终为0，该如何解决呢？并且loss scale一直下降到了0 [deepspeed] OVERFLOW! Rank 0 Skipping step. Attempted loss scale: 1, reducing to 0 [deepspeed] OVERFLOW! Rank 0 Skipping step. Attempted loss scale: 0, reducing to...

[BUG/Help] OVERFLOW导致loss异常

I met the same problem. how to solve it?