Error: PyTorch backend is not built for Docker image

[gswylq]

Summary

Dear all,

When I use the Docker image (deepmodeling/deepmd-kit:latest_cu11) for lmp computation with DPA-3.1-3M.pth, the following error occurs:

RROR on proc 0: DeePMD-kit C API Error: DeePMD-kit Error: PyTorch backend is not built (/project/source/lmp/pair_deepmd.cpp:572)

I hope you can help me.

Thank you very much.
Q Li

DeePMD-kit Version

3.1.0

Backend and its version

ghcr.io/deepmodeling/deepmd-kit:latest_cu11

Python Version, CUDA Version, GCC Version, LAMMPS Version, etc

No response

Details

Dear all,

When I use the Docker image (deepmodeling/deepmd-kit:latest_cu11) for lmp computation with DPA-3.1-3M.pth, the following error occurs:

RROR on proc 0: DeePMD-kit C API Error: DeePMD-kit Error: PyTorch backend is not built (/project/source/lmp/pair_deepmd.cpp:572)

I hope you can help me.

Thank you very much.
Q Li

[gswylq]

The versions of torch and CUDA inside the container (deepmodeling/deepmd-kit:latest_cu11)
do not match the container name. I wonder if this is the issue?

import torch
print(torch.version)
2.3.1+cu121
print(torch.cuda.is_available())
False
print(torch.version.cuda)
12.1

Q L

[njzjz]

RROR on proc 0: DeePMD-kit C API Error: DeePMD-kit Error: PyTorch backend is not built

Hi, this is expected. The PyTorch backend is only enabled in the CUDA 12 image. I cannot make PyTorch work with TensorFlow together with CUDA 11.

The versions of torch and CUDA inside the container (deepmodeling/deepmd-kit:latest_cu11)
do not match the container name. I wonder if this is the issue?

This is indeed a bug. I'll fix it soon.