Can I use GPU with neural-prophet?

[merchen911]

In the code, even though based on the torch, I cant use GPU.

I am using it for the version currently being distributed after personally modifying it.

Hope to know when the neural prophet will support GPU acceleration.

[itay-rubin]

someone?

[wangzhongqing]

Can you save the model file locally? Can you tell me, thank you？

[ourownstory]

currently there is no utility, but you can do it like this:
https://pytorch.org/tutorials/beginner/saving_loading_models.html

Tracked in #27

[wangzhongqing]

Thx,i will try it.

[rishabh1212]

@ourownstory is it possible to use GPU with neural-prophet? or planned for future release??

[ourownstory]

@merchen911 @rishabh1212
Currently, there is no GPU support. We have planned it, but it's likely not soon, as we would first need to achieve better parallelization of the model for this to make a big difference.

However, if you think it may already help you, please feel free to contribute a PR adding GPU support!

[laoliang95]

I would like to ask how NeuralProphet uses GPU

[noxan]

@merchen911 @itay-rubin @wangzhongqing @rishabh1212 We added gpu support based on lightning with the latest release candidate 0.5.0rc1 - happy to hear your feedback and let us know if it works for you - see release notes or the related pull request.

You can either install from source or via pypi with our pre-release version (you have to manually specify exactly 0.5.0rc1 or use the --pre flag.

[hjmok]

hi @noxan , I've installed neuralprophet==0.5.0rc1, and run the following code

trainer_config = {"accelerator":"gpu"}

m = NeuralProphet(
    weekly_seasonality=6,
    daily_seasonality=10,
    trend_reg=1,
    learning_rate=0.01,
    trainer_config=trainer_config
)

df_train, df_test = m.split_df(sf_load_df, freq="H", valid_p=1.0 / 12)

metrics = m.fit(df_train, freq="H", validation_df=df_test, progress="plot")

However, I still run into this error
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!

I have checked that my torch is connected and currently using my NVIDIA GeForce GTX 1650 Ti from this command
print(torch.cuda.get_device_name(0))

Any suggestions?

[noxan]

Hi @hjmok,

@karl-richter is already looking into this, see #1002 and there is meanwhile a fix in review with #1023 - feel free to check out that version and let us know in the issue if it resolves your issue :)