Bug summary
I find property dptest's results to be very different from the learning curve. And I have found the reason. This is a general bug for all heads except for energy head.
In /deepmd/entrypoints/test.py line 127 tmap = dp.get_type_map() if isinstance(dp, DeepPot) else None. If we use DeepProperty or DeepPolar or DeepDOS..... tmap is None. So type_map is not modified again according to type_map in input.json. So atype is wrong in the model forward process. The model prediction value is wrong.
DeePMD-kit Version
All versions
Backend and its version
pytorch
How did you download the software?
Offline packages
Input Files, Running Commands, Error Log, etc.
See above.
Steps to Reproduce
See above.
Further Information, Files, and Links
See above.
Bug summary
I find property dptest's results to be very different from the learning curve. And I have found the reason. This is a general bug for all heads except for energy head.
In
/deepmd/entrypoints/test.pyline 127tmap = dp.get_type_map() if isinstance(dp, DeepPot) else None. If we useDeepPropertyorDeepPolarorDeepDOS.....tmapis None. Sotype_mapis not modified again according totype_mapininput.json. Soatypeis wrong in the model forward process. The model prediction value is wrong.DeePMD-kit Version
All versions
Backend and its version
pytorch
How did you download the software?
Offline packages
Input Files, Running Commands, Error Log, etc.
See above.
Steps to Reproduce
See above.
Further Information, Files, and Links
See above.