Duplicate Storage of Labeled Data During DPGEN2 iteration

I encountered the following issue while running DPGEN2. I used the `dpgen2 download input.json WFID -d collect-data/output/iter_data` code to download the generated training data, but I noticed that there are some duplicate data instances. For example, as highlighted in the red section of the diagram, under the same number of atoms, there are two set.00* folders, and the structural and energy information inside them is completely identical. Additionally, these duplicate data are also being used in subsequent training processes. <img width="840" height="245" alt="Image" src="https://github.com/user-attachments/assets/78b22d13-8780-4dac-bfcf-af78f0bc45d4" /> Is there a way to avoid this issue?