Installing pytorch-lightning on Jetson Nano(ARM)

[enod]

What is your question?

Does pytorch-lightning support ARM processor installation? So far I have failed with pip installation.

What's your environment?

• OS: Linux
• Hardware: Jetson Nano
• Packaging: Pip
• Processor: ARM

[Borda]

I would say that lighting does not have any restrictions n architecture, so it depends on used dependences... could you please try to install from master and pass here error log since not everyone has Jetson Nano(ARM) at his disposal... :]

[enod]

I have managed to install and was able to run MNIST example.

Install pytorch and torchvision via https://devtalk.nvidia.com/default/topic/1049071/pytorch-for-jetson-nano-version-1-3-0-now-available/ which provides prebuilt binary pip wheels.

And install scipy with sudo apt-get install python3-scipy command.

Finally git clone pytorch-lightning and comment out scikit-learn, torch, torchvision in requirements.txt file.
Note that pandas will take at least 30mins.

Hope that helps.

[Borda]

Perfect, happy to hear it!
Btw Commenting out is not needed, since you have these packages they are skipped automatically...

[enod]

idk, for some reason without commenting out, it was still looking for torch, torchvision even though it's installed before.

[neggert]

FWIW, I think we have a loose plan to get rid of the pandas dependency.

[Borda]

just for the record... #693 ^^