OSError: libcusolver.so: cannot open shared object file: No such file or directory

[JunboLu]

Dear all,
I am new user in using gpu4pyscf. I run a small test, but I get an error:

/home/gengzi/bin/deepmd-kit-3.0.0-cuda/lib/python3.12/site-packages/gpu4pyscf/lib/cutensor.py:139: UserWarning: using cupy as the tensor contraction engine.
warnings.warn(f'using {contract_engine} as the tensor contraction engine.')
Traceback (most recent call last):
File "/home/gengzi/WORK/Ln_EMBEDDING/PYSCF/input.py", line 379, in
mf.to_gpu().run(dm)
^^^^^^^^^^^
File "/home/gengzi/bin/deepmd-kit-3.0.0-cuda/lib/python3.12/site-packages/pyscf/df/df_jk.py", line 232, in to_gpu
obj = self.undo_df().to_gpu().density_fit()
^^^^^^^^^^^^^^^^^^^^^^^
File "/home/gengzi/bin/deepmd-kit-3.0.0-cuda/lib/python3.12/site-packages/pyscf/lib/misc.py", line 1513, in to_gpu
import gpu4pyscf
File "/home/gengzi/bin/deepmd-kit-3.0.0-cuda/lib/python3.12/site-packages/gpu4pyscf/init.py", line 17, in
from . import lib, grad, hessian, solvent, scf, dft
File "/home/gengzi/bin/deepmd-kit-3.0.0-cuda/lib/python3.12/site-packages/gpu4pyscf/lib/init.py", line 18, in
from gpu4pyscf.lib import cupy_helper
File "/home/gengzi/bin/deepmd-kit-3.0.0-cuda/lib/python3.12/site-packages/gpu4pyscf/lib/cupy_helper.py", line 25, in
from gpu4pyscf.lib.cusolver import eigh, cholesky #NOQA
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/gengzi/bin/deepmd-kit-3.0.0-cuda/lib/python3.12/site-packages/gpu4pyscf/lib/cusolver.py", line 23, in
libcusolver = ctypes.CDLL('libcusolver.so')
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/gengzi/bin/deepmd-kit-3.0.0-cuda/lib/python3.12/ctypes/init.py", line 379, in init
self._handle = _dlopen(self._name, mode)
^^^^^^^^^^^^^^^^^^^^^^^^^
OSError: libcusolver.so: cannot open shared object file: No such file or directory

I have installed cuda-12.6. How could I solve this problem?
With my best regards,
Junbo

[wxj6000]

Can you check if the installed cuda-toolkit is added to LD_LIBRARY_PATH? GPU4PySCF is based on cupy. You can check if cuda-toolkit can be found through some simple cupy tests.

[cvsik]

From my experience, this error can also happen if you're inside a Conda environment where cuda-toolkit provides only something like libcusolver.so.11 and not libcusolver.so (hard-coded in ctypes.CDLL('libcusolver.so')).

[JunboLu]

Thank you very much! Yes, I use the Conda environment. How could I solve this problem?

…

On Thu, 23 Jan 2025 at 15:41, Maximilian Scheurer ***@***.***> wrote: From my experience, this error can also happen if you're inside a Conda environment where cuda-toolkit provides only something like libcusolver.so.11 and not libcusolver.so (hard-coded in ctypes.CDLL('libcusolver.so')). — Reply to this email directly, view it on GitHub <#308 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AESCXEPPJDHYQRHXMKUK47T2MCMJ3AVCNFSM6AAAAABVME7PFSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDMMBZGA3DKOJRGI> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[wxj6000]

@JunboLu The issue has been fixed in v1.3.1.