Weekly Progress of DeePMD-kit 20220905

[HuangJiameng]

What's new

• docs: make warnings more prominent #1879
• support initilize parameters from a fitting with suffix #1885
• check/update sel for se_atten #1886
• add examples to cli help #1887

Issue

• [BUG] Package conflicts when trying to install deepmd v2.1.4 & cuda 11.6 via conda #1883

Discussion

Answered

• big error in tensile stress computation using dp #1872
  Problem: Lammps simulations run well for a small model of 10675 atoms, while report errors and terminate when the simulation model becomes bigger of 33856 atoms.
  Answer: It was discussed in #1062. It is a bug in CUDA Toolkit 11.1~11.3.

• There is an obvious difference to calcaulate thermal conductivity with different deepmd version #1882
  Problem: Why there is a big difference in the calculation of heat flux results between 2.0.0.b3 and later version.
  Answer: The issue in the calculation of heat flux is fixed since v2.0.1, please see #1093 and #1451

• RDF calculation #1861
  Problem: How to create DFT from deepmd-kit interatom potential
  Answer: We can create RDF directly using deepmd binary data using following code.-- Link - https://github.com/deepmodeling/dpdata/blob/master/dpdata/md/rdf.py

• Can deepmd-kit help me normalize the label values? #1889
  Problem: Can deepmd-kit help me normalize the label values?
  Answer: No, you have to do it manually.

On going

• Question about timestep size in LAMMPS with deepmd models #1884
• Profiling of the DeepMD-kit #1842
• Timeline Trace in Training #1888
• Is it possible to access the Input vector or Last layer of deepmd Neural network? #1852
• dipole model #1893