anhnh2002 / XTTSv2-Finetuning-for-New-Languages

60 stars 17 forks source link

Why is resource usage very low, Does it use gpu or cpu? #13

Closed monitor22 closed 1 month ago

monitor22 commented 1 month ago

Can I let it use more GPU? 2024-27 23_52_40

TugdualKerjan commented 1 month ago

Increase your batch size and watch the magic happen ! It's an empiracal process but quite effiicent ;)

anhnh2002 commented 1 month ago

Can I let it use more GPU? 2024-27 23_52_40

@TugdualKerjan has answered your question :))

monitor22 commented 1 month ago

I have increased the batch size to 128 and its been like this for ~12 hours.

image

anhnh2002 commented 1 month ago

It appears that the training process may not be utilizing the GPU as expected. Here are some steps to diagnose and potentially resolve the problem:

  1. Check GPU availability: Run torch.cuda.is_available() to verify if PyTorch can detect an available GPU.

  2. Verify CUDA installation: Ensure that CUDA is properly installed on your system and that your PyTorch version is built with CUDA support. You can check this by running:

    import torch
    print(torch.__version__)
    print(torch.version.cuda)