Yihe Dong

CUDA error: invalid configuration argument on large

2

Hi, I'm trying to reproduce the pretraining results with multi-GPU training, and encountered this error: `RuntimeError: CUDA error: invalid configuration argument` during validation, but only when the [validation batch size](https://github.com/airsplay/lxmert/blob/master/src/pretrain/lxmert_pretrain.py#L47)...

Do you see a better way to do this without nil checks each time?

Thanks.