Hi! I am wondering if make sense to feed in the complete mini-batch during training at once to the model.
This might increase the utilization of the GPU for RNNs?
Or does pytorch this automatically if it detects a loop around you model?
If so, is there a possibility to convert a a model taking one sequence into a model taking a batch of sequences?
Do you have to split and concat your tensors-flow in your model?
How to archieve that with
nn.LSTM for instance?
Thank you very much!