Skip to content

Fixed Nan in TorchML part + KIM error from requesting virials

  1. Changed the kim_error in kim_cite function to kim_warning in response to dl-poly#2132 (https://gitlab.com/ccp5/dl-poly/-/issues?show=eyJpaWQiOiIyMTMyIiwiZnVsbF9wYXRoIjoiY2NwNS9kbC1wb2x5IiwiaWQiOjE3MzA5MTI1M30%3D)
  2. Moved setting of compute argument pointers in a If branch, after checking whether the model supports the compute argument or not. this removes the error that was getting logged in the kim.log file for partial virials (some models do not support it).
  3. Number of particles for a neighbor list were incorrectly set to number of particle in configuration always. Now it checks if neighbors of those ghost articles are requested, and if yes, sets number of particle = config + ghost atoms. This solves the Nans ML models were reporting earlier. Now i get non-nan numbers for ML models like MACE and Nequip.
Edited by Amit Gupta

Merge request reports

Loading