Page 1 of 1

Errors in ML_MODE=select Training

Posted: Mon Oct 27, 2025 12:23 pm
by qingyu_wang

Dear all,
I encountered an error when setting ML_MODE=select for the merged ML_AB file. My input file, the error report file, and the script file merge.py I used to merge the ML_AB file are provided in the link below.
https://1drv.ms/f/c/b8ec5e7d661b03f9/Eh ... w?e=V33Sp9


Re: Errors in ML_MODE=select Training

Posted: Tue Oct 28, 2025 6:40 am
by ferenc_karsai

It looks like it went through all structures ("2633 done from 2633") and even did the dummy prediction.
Did you get an ML_ABN file and a full ML_LOGFILE?

If yes then it seems like, that the termination of the program has problems. I had one time in the past problems with a faulty mpi that had problems with the termination.

Please also try a different toolchain, but if you have an ML_ABN file than you don't have to repeat the full calculation. I think the results are ok and you only need to test the termination of the code. Use the same ML_AB but trim it to like 10-20 training structures. You also need to set the "The number of configurations" at the beginning of the file to the new number of training structures.