The `force_calculate.py` program gets stuck at the first `forces.in` and cannot continue the loop.

[tululu25]

The problem with this code is that after completing one forces.in calculation, it can't proceed to the next one. It doesn't display the ‘Total wall time’ after the first forces .in calculation is finished, and the GPU memory isn't released, even though the dump has already been generated. My LAMMPS task needs to utilize both the GPU and CPU simultaneously because the potential that i used is DeepMD. How can I fix this?

[tululu25]

If you meet this Question, Some C or Python extension modules or dynamically loaded libraries (such as PyTorch, TensorFlow, OpenCV, etc.) are in an inconsistent state after being forked.
These modules may have dynamically loaded shared libraries in the main process using dlopen and initialized their internal states (such as GPU context, thread pools, etc.). Forking copies these states to the child process, but underlying system resources (such as file descriptors, GPU context) are not actually copied, causing crashes or interface unavailability when the child process makes calls.
To avoid this, I added a setting of multiprocessing module to the main process.Besides that,You may need more memories or GPU memories.

[Tingliangstu]

Hi, thanks so much for your comments