Furkan Kıraç

Problem here is that DataParallel parallelizes data on a specific dimension (which is by default 0). However, RNNs unless you use batch_first=True, chooses the 2nd dimension to be batch. Unfortunately...

Yes, that's the first step. But then you need to adjust internal calculations such that the first dimension is batch.

@mariembenslama It's not very straightforward. I don't know the details of the code. Maybe @Holmeyoung may help you. Best.