deepmd-kit-3.0.3-cuda128-Linux-x86_64.sh lmp without deepmd model

[caijq0712]

Bug summary

Why is the LMP installation I got using the 3.0.3-cuda128 version sh file missing the deepmd module?

DeePMD-kit Version

deepmd-kit-3.0.3-cuda128-Linux-x86_64.sh

Backend and its version

deepmd-kit-3.0.3-cuda128-Linux-x86_64.sh

How did you download the software?

Offline packages

Input Files, Running Commands, Error Log, etc.

Why is the LMP installation I got using the 3.0.3-cuda128 version sh file missing the deepmd module?

Steps to Reproduce

Why is the LMP installation I got using the 3.0.3-cuda128 version sh file missing the deepmd module?

Further Information, Files, and Links

Why is the LMP installation I got using the 3.0.3-cuda128 version sh file missing the deepmd module?

[caijq0712]

I run this version in H800
NVIDIA-SMI 580.95.05 Driver Version: 580.95.05 CUDA Version: 13.0

[njzjz]

@dosu

[dosubot[bot]]

The deepmd-kit offline installer is supposed to provide both the DeePMD-kit and LAMMPS executables, with the deepmd module available in LAMMPS after installation, as long as the environment is set up correctly and dependencies are met (docs). However, there are a few common reasons why the deepmd module might be missing:

1. CUDA 13.0 is not supported by deepmd-kit as of the latest releases, which can cause build failures or missing modules in the binaries (issue).
2. If the conda environment provided by the installer is not activated, LAMMPS may not find the deepmd plugin, resulting in errors like "Unrecognized pair style deepmd" (issue).
3. The deepmd plugin (libdeepmd_lmp.so) must be present in the installation directory (usually under lib/ or lib/deepmd_lmp). If it's missing, the deepmd pair style won't be available (discussion).

Troubleshooting steps:

• Make sure you have activated the conda environment created by the installer before running lmp.
• Check if libdeepmd_lmp.so exists in your installation directory (e.g., $DEEPMD_ROOT/lib/deepmd_lmp).
• If you are running on CUDA 13.0, consider switching to a supported CUDA version (e.g., CUDA 12.x), as CUDA 13.0 is not supported and may prevent the deepmd module from being included in the LAMMPS binary (issue).
• Run lmp -h and check if "deepmd" appears in the list of available pair styles.

If the above steps do not resolve the issue, the missing module is likely due to CUDA 13.0 incompatibility or an incomplete environment setup.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}