The code does not run on a GPU

[mrezasoltani]

The code has some issues when it is run on a GPU. You get CUDA out of memory even with T4 GPU. Plus, there are some other issues such as CUDA error: device-side assert triggered may happen for some values of hyper-parameters.

[SoulProficiency]

sorry!that is a demo.please vist this address:https://github.com/Zhongyang-debug/Sandglasset-A-Light-Multi-Granularity-Self-Attentive-Network-For-Time-Domain-Speech-Separation(finished by anothers in our lab and had been tested in GPU and the results in wsj0(mix) is outperforme than paper's in si-snr criterion)

[mrezasoltani]

Thanks for your reply, however, the link you provided does not work. Can you please check the link?

Thabjs

[SoulProficiency]

following:https://github.com/Zhongyang-debug/Sandglasset-A-Light-Multi-Granularity-Self-Attentive-Network-For-Time-Domain-Speech-Separation
details are being optimized by code author.

[SoulProficiency]

following:https://github.com/Zhongyang-debug/Sandglasset-A-Light-Multi-Granularity-Self-Attentive-Network-For-Time-Domain-Speech-Separation
details are being optimized by code author.

[SoulProficiency]

Sorry for that，we didn't save it. speech separation is not our research topic.what we want is just to test this modeling method and show other people this method is fine.Most details were not taken into account and the training information(such as loss for every epoch) is shown by printf in PyCharm.

[Jasonare]

All right. Since my research topic is speech separation, I would like to save the training logs of various models for comparison. Anyway, thanks for your reply.