Device confusion in ConformerLstmModel - Githubissues

openspeech-team / openspeech

Open-Source Toolkit for End-to-End Speech Recognition leveraging PyTorch-Lightning and Hydra.

https://openspeech-team.github.io/openspeech/

MIT License

677 stars 114 forks source link

Device confusion in ConformerLstmModel #125

Open Stasiche opened 2 years ago

Stasiche commented 2 years ago

Environment info

Platform: Ubuntu 20.04
Python version: 3.9
PyTorch version (GPU?): pytorch-lightning==1.5.2
Using GPU in script?: No

Information

Model I am using (ListenAttendSpell, Transformer, Conformer ...): ConformerLstmModel

I hadn't got enough memory to use GPU, but I wanted to try one thing. So I set "trainer=cpu". I got "RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument index in method wrapper__index_select)" on https://github.com/openspeech-team/openspeech/blob/main/openspeech/decoders/lstm_attention_decoder.py#L140

As I can see the reason is that it's not safe to do this this way: https://github.com/openspeech-team/openspeech/blob/main/openspeech/decoders/lstm_attention_decoder.py#L243 Because my card is still installed, so torch.cuda.is_available() == True. But my model is on CPU.

sooftware commented 2 years ago

Thank you for letting us know. We'll check. cc. @upskyy