Using the DP model to predict the energy of each frame structure in the LAMMPS trajectory

### Summary

I now have a lambps trajectory that contains many structures, but these structures are not the original lambps trajectory structure, but some extracted molecular clusters. Therefore, the number of atoms or types of elements in each frame may be different, but overall they are several elements contained in the original trajectory. But for this manually spliced lambps trajectory, the model's predicted code will report an error. One method is for me to extract each molecular cluster into a separate lambps trajectory, write a script to read each file, and achieve energy prediction, but the efficiency is too slow because I have tens of thousands of molecular cluster structures.

### DeePMD-kit Version

2.1.5

### TensorFlow Version

2.8.0

### Python Version, CUDA Version, GCC Version, LAMMPS Version, etc

_No response_

### Details