How can I convert CP2K MDtrajectories into the format used for training with DeepMD-kit?

[naturec1-a]

Now, I have constructed 400 randomly generated periodic cells containing randomly placed and oriented small molecules including 30 kinds of organic molecules with random composition, the number of atoms of these simulation systems is among the range of 200-500. I use the following codes " data = dpdata.LabeledSystem(task_path, cp2k_output_name=cp2k_output_name, fmt='cp2k/aimd_output')" to draw information.
When using the above codes for conversion, it will extract a whole frame of the MD trajectory as a conformation: for example, a simulation system contains 200 atoms, then after extraction, each point contains all 200 atoms in the whole frame, and its coordinates contain 600 dimensions. Therefore, I have the following two questions:

1. Is this extraction method correct? Should I take a whole frame on the MD trajectory as a point or extract some clusters from the whole frame (i.e., recalculate the single point energy of some molecules) for DeePMD training?
2. Most of the AIMD trajectories are vibrations of some bonds, which means there will be a large number of repeated conformations, so they need to be screened and deduplicated, otherwise the training results will not be good if there are too many repeated conformations. How should this be performed?