Weekly Progress of DeePMD-kit 20221010

[HuangJiameng]

What's new

• search TF deps from LD_LIBRARY_PATH; eliminate errors #1976
• find protobuf headers in an automatic way #1975
• support bfloat16 #1970
• fix googletest download #1949
• enable gcov for Python tests #1969
• fix typo in the doc of loc_frame #1963
• Upload the finetune api #1946
• fix compatible with importlib_metadata 5.0.0 and Python 3.12 #1961
• use CPP_CXX_ABI for Python libraries #1960

Issue

• [BUG] Type map for single element #1931
• Negative volume detected for some simulation cells during training #972
• [Feature Request] set LAMMPS_VERSION_NUMBER before lammps make #1947

Discussion

• [LAMMPS] Error while doing simulation #1968
  If you find:

1. ERROR: Non-numeric pressure - simulation unstable (src/fix_nh.cpp:1059) Last command: run 100000
2. Sometimes Lost Atom Error
   It happens when the model is not good enough to cover the structures and makes the simulation crash.

• Bfloat16 #1933
• Installing DeePMD-kit C++ Interface #1974
• Timeline Trace in Training #1888
• Large Dataset Requirement For Analysis #1964
  Please note that the training the dataset from https://dplibrary.deepmd.net/ for Water and Copper does not "comprising millions of atoms". We trained DP models from dataset of dozens of atoms, and did MD simulations of millions of atoms.
• Info About New Features Still Need to be Added To This Framework #1831
• Large RMSE errors for forces #1972
• Running Lammps parallel in conda environment installation (easy installation method) #1978