Trying to do classification with QRNN leads to a model input size mismatch between the previous input which has had its padding stripped and the current input which is still padded
Error
invalid argument 0: Sizes of tensors must match except in dimension 1. Got 2 and 1 in dimension 0 at /opt/conda/conda-bld/pytorch_1573049306803/work/aten/src/THC/generic/THCTensorMath.cu:71
Replication
The error can be replicated with the ULMFiT tutorial in nb38 where the only change made is swapping the architecture and config for the learner:
As outlined on the forums: https://forums.fast.ai/t/fastai-v2-chat/53518/673
Trying to do classification with QRNN leads to a model input size mismatch between the previous input which has had its padding stripped and the current input which is still padded
Error
Replication
The error can be replicated with the ULMFiT tutorial in nb38 where the only change made is swapping the architecture and config for the learner:
Language leaner (trains as expected):
Classifier learner (raise error when trained)
Full stack trace