MAGMA not initialised during training

[wallematthias]

Hi,

when I train the dbsr model on the default_realworld setting I get the following error during training:

Error in magma getdevice arch: MAGMA not initialized (call magma init() first) or bad device
First, I thought this was due to multiprocessing start method, but it is already set to spawn.

Any Ideas?

Best,
Matthias

[yjsunnn]

What's your pytorch version? When my torch version==1.9.0, I got the same MAGMA error. And I fixed the problem by downgrading pytorch to 1.7 version.

[wallematthias]

Hi susan,

You are correct I had 1.9 installed, downgrading to the following solved the problem:

# CUDA 10.2
conda install pytorch==1.7.1 torchvision==0.8.2 torchaudio==0.7.2 cudatoolkit=10.2 -c pytorch

THANK YOU :)

[goutamgmb]

Hi,

Thanks for raising this issue @wallematthias and for the solution @susan-sun1999. I will check this and post a note on the install page.