[BUG] Error while training DPLR

[siddarthachar]

Bug summary

Error while training a DPLR model using a deepdipole model. I tried using the example for water in https://github.com/CSIprinceton/workshop-july-2023/tree/main/hands-on-sessions/day-3/8-deep-potential-long-range/1-train to make a deepdipole first and then a DPLR. The deepdipole was trained as per: https://github.com/CSIprinceton/workshop-july-2023/tree/main/hands-on-sessions/day-3/7-deep-wannier/2-train_dipole_model and it worked fine. I even tested these models to predict the atomic dipole moment vectors for water. However, training the DPLR gave this error:

NotImplementedError: Cannot convert a symbolic Tensor (strided_slice_3:0) to a numpy array. This error may indicate that you're trying to pass a Tensor to a NumPy call, which is not supported 

DeePMD-kit Version

2.0.3

TensorFlow Version

2?

How did you download the software?

conda

Input Files, Running Commands, Error Log, etc.

Input file: https://github.com/CSIprinceton/workshop-july-2023/blob/main/hands-on-sessions/day-3/8-deep-potential-long-range/1-train/input.json

Running Commands: dp train input.json

Error log:
slurm-624716.txt

Steps to Reproduce

Use the files in https://github.com/CSIprinceton/workshop-july-2023/tree/main/hands-on-sessions/day-3/ to reproduce .pb files for DPLR training.

Further Information, Files, and Links

No response

[Yi-FanLi]

Hi Siddarth,

Thanks for reporting this issue. I am so happy that you reach out to us after you go back from our workshop.

I have tried to reproduce your issue in my own environment. However, the training runs correctly on my side.

From the information given in the issue, I do not have a clear idea on the reason of your issue. However, I suspect that the issue is because you are using an inappropriate environment. Specifically, it might be due to that the tensorflow version is improper. I suggest that you might create a new conda environment and install the newest release of deepmd-kit from the deepmodeling conda channel: https://anaconda.org/deepmodeling/deepmd-kit.

Best,
Yifan

[siddarthachar]

Hi Yifan

It's nice to hear back from you. I hope things are going well!

I did install the latest version and it seems to be working now. I had a followup question about selecting the system charge and the model charge map. I have a system with C, O, H, so the system charge would just be [4, 6, 1] and the model charge map should be [-8, -8], right? The first is the just the number of valence electrons and the second is -2*the number of MLWC for each element type. Is this assumption correct?

Best,
Siddarth

[njzjz]

I'll close this issue since there is nothing wrong with the software.

@Yi-FanLi will answer the question.