-
Notifications
You must be signed in to change notification settings - Fork 74k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
No Module Named '_pywrap_tensorflow_internal' (still without working solution) #11571
Comments
I'm getting the same error, with a different environment (Tensorflow r1.2, on a MacBook with OSX 10.12.5 and no GPU support). The installation works, but I'm getting errors, when I try to import TF in python.
I'm working in a
I tried to install the default pip package using Related issues/questions (Github/Stackoverflow) and their answers pointed out problems with CUDA or missing Windows dll files, which do not seem to fit here. |
@StephanH84 This is a common error that's usually related to missing DLLs and your system Path not being configured correctly. Also in your issue filing you say that you've installed from source but then list a nightly binary and say you've not compiled from source. Can I assume that you installed using a binary via pip? If so may I suggest uninstalling and installing straight from Pypi Let us know how you get on. Good luck! @pjaehrling It might be best that you file a Stack Overflow issue as your issue appears to be separate (although similar!) due to the different operating system, different version of tensorflow. I have no experience with the Mac version so hopefully someone else will come and help you! |
@mrry should we check in the script? |
@jubjamie: I already posted my problem on stackoverflow, unfortunately I got no response yet. |
For me it was possibly solved by adding Visual Studio directories to Windows' path variable (C:\Program Files (x86)\Microsoft Visual Studio\Shared\14.0\VC\bin\amd64, C:\Program Files (x86)\Microsoft Visual Studio\Shared\14.0\VC\bin) and then reinstalling tensorflow-gpu via pip (but here it took a version from cache):
There were some issues at the start as comments:
Apart from that at least for my issue "this issue" can be closed. |
Those are just warnings to do with possible ways to improve performance if compiling from source. They are normal and can be ignored. If you're happy the issue is solved you can close it yourself! |
@jubjamie should I open another issue? I couldn't find a way to make it work nor got a response to my Stack Overflow issue so far. I also tried to install tensorflow outside of the virtualenv and tried to build a pip wheel from source for the latest release version 1.3, but got the same error. When I build the
|
You're best off opening a separate issue to @StephanH84 if you have a different problem. In the new issue be sure to fill out the full template and also link and Stack Overflow questions you're already filed. |
I just wrestled with this for a couple hours (Windows 10, Anaconda 4.4.0 64 bit). I had been trying to install tensorflow-gpu using pip with no success. I followed all the guides I could find, tried using Python 3.5.3, installed cudnn 5.1 and 6.0, tried different combinations using conda environments, and still ran into the same error. I found that installing via conda rather than pip worked the first time. Conda installed cudatoolkit, cudnn, libprotobuf, and protobuf and updated itself and vs2015_runtime. Apparently conda is better at putting all this stuff in the right place than I am. Praise conda. |
Had the same problem. Resolved using CUDA 9.0 and tf-nightly-gpu as suggested above. I have put up instructions here: https://github.com/rohit-patel/Install_Instructions-Win10-Deeplearning-Keras-Tensorflow |
I meet this issue too. I installed anaconda on my Windows10 system. I tried to install TF through conda channel, it was OK at that time, however, the version of TF was 1.2.1, which I thought it was too old, besides, the instructions on tensorflow.org said TF installed via conda channel is not supported officially. So I switched to the official installation via the command: pip install tensorflow, however, the issue raised up. |
Windows 10, conda env, |
System information
Yes, it is one general import command (see below).
Windows 10
From source (nightly build tensorflow_gpu-1.2.1-cp35-cp35m-win_amd64.whl from 2017/07/13).
1.2.1 (command does not work, it includes the failed command)
3.5.2
Not compiled from source.
CUDA 8.0, cudnn64_5.dll (Windows file description is: NVIDIA CUDA CUDNN Library. Version 8.0.54)
GPU NVidia Geforce 1050
import tensorflow
Describe the problem
Importing TensorFlow causes an error:
Source code / logs
check_tensorflow.py produces:
The issue was already reported to Stackoverflow:
Stackoverflow issue
The text was updated successfully, but these errors were encountered: