search

thorben-frank / mlff

Build neural networks for machine learning force fields with JAX
MIT License
56 stars 13 forks source link

error at the beginning of the training #25

Open nimakt7192 opened 4 months ago

nimakt7192 commented 4 months ago

Hi

I'm trying to a train a model with So3krates but I'm getting an error (I attached the log file). I tried to run the example that is already there (ethanol) and it failed. Then I tried my own data, following the instructions and got exactly the same error. The error seems to stem from orbax but I'm not sure how to fix this. Do you have any suggestions? error.log

Best, Nima

thorben-frank commented 3 months ago

Hi Nima,

thanks for raising the issue. With my current settings I can not reproduce the error. I can remember that this error appeared for me at some point as well, but I can not remember the specific fix unfortunately.

I assume you have already tried passing an absolute path as --ckpt_dir? Are you running train_so3krates or train_so3kratACE, your logs imply the ladder I guess? In case this is true, does the error also appear for train_so3krates?

Could you maybe share the specific version of orbax you are using as well as the python version?

Best, Thorben