search

atomistic-machine-learning / schnetpack

SchNetPack - Deep Neural Networks for Atomistic Systems
Other
774 stars 214 forks source link

Problems during training #613

Closed MasterLucas closed 6 months ago

MasterLucas commented 7 months ago

Dear SchNetPack developers,

I am facing some problems training SchNet on the QM9 dataset according to the instructions in the docs. When I use the specified command "_$ spktrain experiment=qm9atomwise", the following error is raised:

TypeError: Cannot convert a MPS Tensor to float64 dtype as the MPS framework doesn't support float64. Please use float32 instead.

Could you please help me with this?

Yours faithfully,

jnsLs commented 7 months ago

Dear Lucas,

seems like you are using a MAC. If you don't have the option to run SchNetPack on a computation cluster or a cloud service, the easiest option should be disabling your GPU and train on the CPU instead.

Best, Jonas

MasterLucas commented 6 months ago

Dear Jonas,

Thank you very much for your reply and help. I managed to run SchNetPack both in the cloud and on a supercomputer. However, I am still a bit confused about what happened. I am using a Mac, but there is no GPU in it. Could you think of another reason for this error and how I could solve it?

Yours sincerely,

Lucas Bandeira

jnsLs commented 6 months ago

Hi Lucas,

have you tried adding the argument trainer.accelerator=cpu ?

MasterLucas commented 6 months ago

Now it works!

Thank you very much, Jonas!