Error when training a model on CUDA

[stefan-it]

Hi,

I think this is a regression bug that was introduced after merging the CUDA semantics pull request #402.

Here's an error message that comes from the test_model_integration.py integration test:

>               tag_tensor = torch.LongTensor(sentence_tags)
E               TypeError: expected torch.LongTensor (got torch.cuda.LongTensor)

I got a similar error message when I tried to test my recent ELMo transformer embeddings on GPU.

To replicate the error just run the integration tests (pytest --runintegration) on a machine with GPU/CUDA support :)

[stefan-it]

Thanks, it is working now :)