How to use different strategies between training and validation loops?

[Meehai]

Hello, I'd like to do the following workflow and was wondering if it's possible with the current pl.Trainer:

• train set using DDP (on >=2 GPUs) -> dataset is split and metrics are shared and synced using self.log(..., sync_dist=True) in the on_train_batch/epoch_end methods
• validation set using no strategy, computed on the first GPU only -> full validation set remains in GPU 0

However, as far as I know, the trainer object receives a single strategy parameter (i.e. there's no validation_strategy parameter). Is there any way to achieve what I want?