Roman Kempt

I had the same issue and forcing a reinstall of `pybind11-global` via conda-forge seems to have fixed it (conda install -c conda-forge pybind11-global). But I have a hard time reproducing...

All these files exist in a directory path like this: ... lib/python3.9/site-packages/tensorflow/include/tensorflow/core/framework/ ... lib/python3.9/site-packages/tensorflow/include/tensorflow/core/public/ ... lib/python3.9/site-packages/tensorflow/include/tensorflow/core/platform/ But they are not found by CMake. Might be related to this discussion https://github.com/deepmodeling/deepmd-kit/discussions/272,...

Sorry for the delay! TensorFlow has been preinstalled by the HPC support (I've struggled to install tensorflow myself with NCCL etc.). The google protobuf directory is missing, the only protobuf...

The output of this is: ``` ldd /p/software/juwelsbooster/stages/2022/software/TensorFlow/2.6.0-gcccoremkl-11.2.0-2021.4.0-CUDA-11.5/lib/python3.9/site-packages/tensorflow/libtensorflow_framework.so.2 linux-vdso.so.1 (0x00007fff56f58000) libdl.so.2 => /usr/lib64/libdl.so.2 (0x000014828c7d0000) libm.so.6 => /usr/lib64/libm.so.6 (0x000014828c44e000) libprotobuf.so.3.17.3.0 => /p/software/juwelsbooster/stages/2022/software/protobuf/3.17.3-GCCcore-11.2.0/lib/libprotobuf.so.3.17.3.0 (0x000014828c14c000) libsnappy.so.1 => /p/software/juwelsbooster/stages/2022/software/snappy/1.1.9-GCCcore-11.2.0/lib/libsnappy.so.1 (0x000014828deef000) libdouble-conversion.so.3 => /p/software/juwelsbooster/stages/2022/software/double-conversion/3.1.6-GCCcore-11.2.0/lib/libdouble-conversion.so.3 (0x000014828dedd000)...

Thank you for responding so quickly! The current devel branch raises the following error: ``` -- Found CUDA: /p/software/juwelsbooster/stages/2022/software/CUDA/11.5 (found version "11.5") -- Found CUDA in /p/software/juwelsbooster/stages/2022/software/CUDA/11.5, build nv GPU...

Thank you for all your work! I've attached the LD_DEBUG log and the LD_LIBRARY_PATH is at the end of the file. [LD_DEBUG.txt](https://github.com/deepmodeling/deepmd-kit/files/9744142/LD_DEBUG.txt) With #1976, I get the following error: ```...

The error persists, unfortunately. I've modified the Findtensorflow.cmake to write out the libraries that CMake detects as linked to tensorflow: ``` -- UNRESOLVED_DEPENDENCIES_VAR -- libdouble-conversion.so.3 -- libjpeg.so.8 -- libprotobuf.so.3.17.3.0 --...

I can't modify the tensorflow module unfortunately, since its provided by the HPC. The issue with the protobuf might also boil down to some reading/access rights, since only the libraries...

hm, the `pyproject.toml` looks like this, if you are referring to that: ``` [build-system] requires = [ "setuptools>=42", "wheel", "scikit-build", "cmake", "pybind11", "numpy", "spglib", "ninja; platform_system!='Windows'" ] build-backend = "setuptools.build_meta"...

I have the same issue (see here https://stackoverflow.com/questions/70711640/c-program-compiles-and-links-but-then-raises-undefined-symbol-error). I can't reproduce it on my own machine, but it happens for colleagues every time. Gives me quite a headache by now