deepmodeling
[BUG] Changing `nlayer` lead no error report while fine-tuning
Bug summary
When changing the nlayer parameter in input.json while fine-tune the DPA310-3M pre-trained model, deepmd-kit will throw no error and process the training for the model of this nlayer. I have no idea that it is correct or wrong. In my knowledge the nlayer should keep the same.
DeePMD-kit Version
3.1.0
Backend and its version
PyTorch
How did you download the software?
Offline packages
Input Files, Running Commands, Error Log, etc.
input is the same as any dpa3-fine-tuning input.json, only the nlayer parameter is changed
Steps to Reproduce
change the nlayer of input.json for dpa3 fine-tuning, and perform this fine-tuning
Further Information, Files, and Links
No response
The configuration of the descriptor in input.json is dropped. There should be a clear message, though, if there is not.
It seems that now only the missing keys in the checkpoint will through an error, but not for the extra keys (it is the case if you set a smaller nlayer). A warning should be added.
