Does neuralforecast require GPUs? Or it is possible to run it on CPUs at slower rate?

you can use CPUs, up to 10 i believe

Just curious, why only 10?

no idea i just read that in the docs

Hey @Valeriy, you can use GPUs and CPUs It is very easy to switch modes in Google Colab https://github.com/Nixtla/neuralforecast/blob/main/nbs/examples/Getting_Started.ipynb You can try CPUs, TPUs and GPUs

Thanks @Kin Gtz. Olivares I am aware that one can change it. Was just asking because didn’t manage to run on CPUs, basically not much was happening and it was not possible to say if the model was training or not. Will try again.

In general, I would advice against CPUs, because it takes too much time and normally the Google Colab subscription of 10 USD monthly was enough for semi heavy training jobs

Yes, we are trying to put into production in azure and considering if it is worth paying for GPUs as there doesn’t not seem to be easy way to integrate with Colab. For R&D Colab was fine indeed

For training purposes GPU is good, for inference purposes with a pretrained model you might be able to use CPU

I’ve used it on cpu on windows and on linux (docker) without a problem.