Error running dp_train using dp_train

[k50112113]

Hi:

I have installed tensorflow r1.8 and DeePMD r0.11 on cluster and I am currently trying to train a model.
It works well when I am testing on home node.
But error below occurs when I was trying to submit jobs using PBS scripts:

File "/usr/local/anaconda/5.2.0/3/lib/python3.6/site-packages/tensorflow/python/pywrap_t
ensorflow.py", line 58, in
from tensorflow.python.pywrap_tensorflow_internal import *
File "/usr/local/anaconda/5.2.0/3/lib/python3.6/site-packages/tensorflow/python/pywrap_t
ensorflow_internal.py", line 28, in
_pywrap_tensorflow_internal = swig_import_helper()
File "/usr/local/anaconda/5.2.0/3/lib/python3.6/site-packages/tensorflow/python/pywrap_t
ensorflow_internal.py", line 24, in swig_import_helper
_mod = imp.load_module('_pywrap_tensorflow_internal', fp, pathname, description)
File "/usr/local/anaconda/5.2.0/3/lib/python3.6/imp.py", line 243, in load_module
return load_dynamic(name, filename, file)
File "/usr/local/anaconda/5.2.0/3/lib/python3.6/imp.py", line 343, in load_dynamic
return _load(spec)
ImportError: libcuda.so.1: cannot open shared object file: No such file or directory

But I did not build my tensorflow with CUDA support and it seems that it is not using the tensorflow I built (I am installing tensorflow's C++ interface)
It seems that dp_train only works with tensorflow's python interface?

Thanks for your time,
Lee, Shao-Chun

[jameswind]

Yes, Shao-Chun, dp_train only requires tensorflow's python interface. Most likely you've installed a python interface of tensorflow's with a GPU support. Can you double-check? If you have available GPU machines, you may change the PBS submission script, require a single GPU, and try again. If not, you may install tensorflow's python interface with CPU support only. Best, Linfeng

…

On Mon, Oct 14, 2019 at 1:07 PM k50112113 ***@***.***> wrote: Hi: I have installed tensorflow r1.8 and DeePMD r0.11 on cluster and I am currently trying to train a model. It works well when I am testing on home node. But error below occurs when I was trying to submit jobs using PBS scripts: File "/usr/local/anaconda/5.2.0/3/lib/python3.6/site-packages/tensorflow/python/pywrap_t ensorflow.py", line 58, in from tensorflow.python.pywrap_tensorflow_internal import * File "/usr/local/anaconda/5.2.0/3/lib/python3.6/site-packages/tensorflow/python/pywrap_t ensorflow_internal.py", line 28, in _pywrap_tensorflow_internal = swig_import_helper() File "/usr/local/anaconda/5.2.0/3/lib/python3.6/site-packages/tensorflow/python/pywrap_t ensorflow_internal.py", line 24, in swig_import_helper _mod = imp.load_module('_pywrap_tensorflow_internal', fp, pathname, description) File "/usr/local/anaconda/5.2.0/3/lib/python3.6/imp.py", line 243, in load_module return load_dynamic(name, filename, file) File "/usr/local/anaconda/5.2.0/3/lib/python3.6/imp.py", line 343, in load_dynamic return _load(spec) ImportError: libcuda.so.1: cannot open shared object file: No such file or directory But I did not build my tensorflow with CUDA support and it seems that it is not using the tensorflow I built (I am installing tensorflow's C++ interface) It seems that dp_train only works with tensorflow's python interface? Thanks for your time, Lee, Shao-Chun — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <https://github.com/deepmodeling/deepmd-kit/issues/99?email_source=notifications&email_token=AEJ6DC5QIV54YT5FSV6UOFDQOSRM3A5CNFSM4JARYVTKYY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4HRUYDAA>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEJ6DC7FHRQILZXVE3VSOELQOSRM3ANCNFSM4JARYVTA> .

[k50112113]

Thanks, I'll reinstall my tensorflow with Python Interface and see if it works.