DP+LAMMPS Precision

[npiroozan]

Hello,

I know this has been discussed before but the solution is a bit unclear to me. The following is the issue:

tensorflow/core/framework/tensor.cc:725] Check failed: dtype() == expected_dtype (1 vs. 2) double expected, got float

Clearly the model has been trained for single precision but LAMMPS is expecting inference output of double precision. I'm using version 2.1.2 of DP-kit and build DP+LAMMPS with the plugin mode.

How can this be resolved?

Thank you very much!

Warm Regards

[njzjz]

See #1907, if you installed lammps from our conda channel instead of the old lammps-dp package.

[njzjz]

The following files should be present:

• lib/libtensorflow_framework.so
• lib/libtensorflow_cc.so

I think you can manually check it.

[npiroozan]

I do not see it there. The steps that I took to build tensorflow are as follows:

1. Create conda environment
2. Pip Install Python 3.9
3. Pip Install tensorflow 2.9
4. Install DeePMD
5. Set $tensorflow_root and link with DeePMD/source/build

Is there an additional step?

[njzjz]

C++ interface is expected as described here.

If one wants to use the Python interface, its files must be adjusted to the same as the C++ interface.

[npiroozan]

Understood, so positioning of files in directory pathway is the problem it seems?

[npiroozan]

Everything is working well now, thank you for your help!