Weekly Progress of DeePMD-kit 20220822

[HuangJiameng]

What's new

DeePMD-kit v2.1.4 release

Highlights

Enhancements:

• add core api docs #1800
• add op docs #1804
• There's no need for building libtensorflow_cc.so anymore #1744

Bug fixings:

• revert prod_force OMP in #1360  #1862
• fix typo in hip assert error message #1802
• fix memory leaking of GraphDef #1811

Check out https://github.com/deepmodeling/deepmd-kit/releases/tag/v2.1.4 to learn more about DeePMD-kit v2.1.4.

Discussion

• in what cases must fp64 be used to train the model #1860
  The author is wondering in what cases must fp64 be used to train the model in order to get enough accuracy. Awaiting answers.

• Doubts about free energy calculation of ice #1859
  If you find the calculated free energies abnormal, you may follow the section 9 of this work http://iopscience.iop.org/0953-8984/20/15/153101 to check the calculation of free energies.

• UnsatisfiableError when I install Deepmd-kit 2.1.1 #1761
  UnsatisfiableError occurred when installing DeePMD-kit 2.1.1(gpu version). This problem is said to exist in DeePMD-kit 2.1.3 too.

• Inference #1822

1. One can output the Tensorboard trace for inference by writing a sample Python script with input data and use the TensorFlow Profiler to trace the bottlenecks.
2. dpdata can help you get the Radial Distribution Function of your model.
   RDF calculation #1861 and Profiling of the DeepMD-kit #1842 are also related to this topic.

• deepmd-kit package is inconsitent with TensorFlow Runtime #1869
  One may check the packages (conda list, pip list) and environment variables (env) to see if anything is wrong if RuntimeError This deepmd-kit package is inconsistent with TensorFlow Runtime occurs.

[wanghan-iapcm]

We may want to mention the release of v2.1.4