YueWen

Results 2 issues of YueWen

### Is there an existing issue for this? - [X] I have searched the existing issues ### Current Behavior 使用PTV2训练方式,修改train.sh中的CUDA_VISIBLE_DEVICES=0,1,在训练开始时就报错,提示OOM,请问大家是怎么解决的?两张4090 ### Expected Behavior _No response_ ### Steps To Reproduce ###...

使用以下代码,进行单机多卡训练,出现两张卡比一张卡训练速度慢, trainer = pl.Trainer.from_argparse_args(args, callbacks=[checkpoint], gpus=2, logger=False, accelerator='dp') 请问是哪里设置的有问题吗? 单卡训练: ![image](https://github.com/renmada/t5-pegasus-pytorch/assets/96854562/abd454e5-51c5-4e05-ad8c-a025df5ef157) duo'ka 多卡训练: ![image](https://github.com/renmada/t5-pegasus-pytorch/assets/96854562/a0b6868f-002a-47bf-b42f-baa6a4323aae)