Nixtla's Open Source Time Series Ecosystem.

Nixtla Community

Hello!  :slightly_smiling_face: I'm running the <https://github.com/Nixtla/neuralforecast> example.  I notice it's using CPU... how can I configure to utilize GPU?  (Cuda 11.6)

image.png

Hi <@U05H1417JDU>! You do not need to change the code in the examples to use GPU! Torch automatically determines when there is a GPU with cuda available and will use it. Where are you running the experiments? If you are in Google Colab, you need to change the instance to one with GPU.

For other cases, you need to make sure that the installation was done properly so that `cuda` is available for your PyTorch installation. Installing directly with `pip` or `conda` should install everything correctly. You can check that everything works properly with `torch.cuda.is_available()` , it should return True

<@U05H1417JDU>, if `torch.cuda.is_available()` returns true, then GPU should be used automatically by NeuralForecast. You don't have to configure anything extra.

Thank you both for the info!  It's working now, I had to explicitly install cuda toolkit for Torch, I thought that was included by default.