Cong Luwen

Results 4 comments of Cong Luwen

maybe decrease the batch size will solve the out of memory issue?

@jamiechoi1995 add `"verbose_beam":0` in the `eval_kwargs`, this will not print the `done_beams` which may solve this issue. If you do want to print the `done_beams`, try to pass the `dp_model.module`...

I think this is because my poor implementation. I did not do any optimization on the code. So It's far more slowly than the original code(which is written in c++)....