[tensorflow] On Windows, running "import tensorflow" generates No module named "_pywrap_tensorflow" error

On Windows, TensorFlow reports either or both of the following errors after executing an import tensorflow statement:

• No module named "_pywrap_tensorflow"
• DLL load failed.

After much trial and error, and making sure VC++ 2015 Redistributable, cuDNN DLL and all other dependencies are accessible from PATH, it looks like Tensorflow GPU works only with Python 3.5.2 (as of this writing)

So if you're using Anaconda

• conda create -n tensorflow-gpu python=3.5.2
• activate tensorflow-gpu
• pip install tensorflow-gpu

Then open the python interpreter and verify

>>> import tensorflow as tf
>>> sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))

Found device 0 with properties:
name: GeForce 940M
major: 5 minor: 0
memoryClockRate (GHz) 1.176
pciBusID 0000:06:00.0
Total memory: 2.00GiB
Free memory: 1.66GiB

Credits: this neat guide

In my case the "cudnn64_6.dll" file in the /bin folder had to be renamed to "cudnn64_5.dll" for the error to go away. I easily spent two hours to figure this out, and I followed the official install guide to the letter. This is true for installation via pip (officially supported) and conda (community supported).

Dll not found. Install Visual C++ 2015 redistributable to fix.

TensorFlow requires MSVCP140.DLL, which may not be installed on your system. To solve it open the terminal en type or paste this link:

C:\> pip install --upgrade https://storage.googleapis.com/tensorflow/windows/cpu/tensorflow-1.0.0-cp35-cp35m-win_amd64.whl 

Note this is to install the CPU-only version of TensorFlow.

My two cents:

I had a ton of problems trying to get my CUDA 8.0 installed properly on Windows 7. I had a previous version installed and I wanted to upgrade so I uninstalled it and tried to install CUDA 8.0 (for tensorflow 1.3). The installation failed every single time, I tried to downgrade to CUDA 7.5 and was able to install it but had a ton of problems with tensorflow (similar to the PATH problem described here). Long story short: what worked for me was:

1) Uninstall EVERY NVIDIA component (except the display graphics driver)

2) Download CUDA toolkit 8.0 (and the patch) https://developer.nvidia.com/cuda-downloads

3) Check the CheckSum MD5 (I used MS https://www.microsoft.com/en-ca/download/confirmation.aspx?id=11533 but any would do) to make sure they were OK (it happened several times that the installer was not dowloaded properly because my WiFi router apparently).

4) Run the CUDA toolkit installer as root

5) download the cudnn 8.0 v6 and add its location to the PATH variable https://developer.nvidia.com/rdp/cudnn-download

Hope that helps and saves some headaches...

NOTE: This script helped me a lot to debug the problem! (Thank you mrry) https://gist.github.com/mrry/ee5dbcfdd045fa48a27d56664411d41c

In case you are trying to install tensorflow GPU in Windows, you can find this easy interesting tutorial.

Note: If you are using PyCharm for example, you have to change the interpreter to the created conda environment.

tensorflow 1.3 does not support cuda 9.0 yet. I degrade to cuda 8.0, then it works.

I posted a general approach for troubleshooting the "DLL load failed" problem in this post on Windows systems. For reference:

1. Use the DLL dependency analyzer Dependencies to analyze <Your Python Dir>\Lib\site-packages\tensorflow\python\_pywrap_tensorflow_internal.pyd and determine the exact missing DLL (indicated by a ? beside the DLL). The path of the .pyd file is based on the TensorFlow 1.9 GPU version that I installed. I am not sure if the name and path is the same in other TensorFlow versions.

2. Look for information of the missing DLL and install the appropriate package to resolve the problem.

I will try to give the solution that worked for me. It seems that different set of problems can lead to this situation.

32 bit software works in 64 bit OS. I installed anaconda-3 (32 bit) in my 64 bit OS. It was working perfectly fine. I decided to install tensorflow in my machine and it wouldn't install at first. I was using conda environment to install tensorflow and got this error.

Solution is if you are running 64 bit OS, install 64 bit anaconda and if 32 bit OS then 32 bit anaconda. Then follow the standard procedure mentioned in tensorflow website for windows (anaconda installation). This made it possible to install tensorflow without any problem.

I have Win7 Pro 64-bit on AMD cpu, no gpu. I was following the instructions under "Installing with native pip" at https://www.tensorflow.org/install/install_windows. The installation step went ok but the attempt to import tensorflow produced the infamous:

ImportError: No module named '_pywrap_tensorflow_internal'

This seems to be one of those situations where a lot of unrelated things can go wrong, depending on configuration, which all cascade through to the same error.

In my case, installing MSVCP140.DLL was the answer.

You have MSVCP140.DLL already if

1. you have a file C:\Windows\System32\MSVCP140.DLL, AND
2. if you have a 64 bit system, then you additionally have C:\Windows\SysWOW64\MSVCP140.DLL.

I installed it manually, which was unnecessary (the redistributable is not the whole Visual C++ development mess and isn't large). Use the link posted earlier in this thread to install it: Visual C++ 2015 redistributable.

Also, I recommend that you override the default install directory for Python and put it anywhere not under C:\Program Files, because Windows tries to write-protect files there, which causes problems later.

The problem for me was the cuDNN library which didn't match the requirements of the graphics card. I downloaded the 6.0 version but for my GTX980ti but the recommended compute capability on the nvidia website was 5.1 ( http://developer.nvidia.com/cuda-gpus ) so I downloaded 5.1 and replaced the 6.0 version and as soon as I've done that it started working.

Ran into the same problem (in 2019/09/09) while investigating [SO]: Error while training using the estimator API in tensorflow.

Setup:

• Win 10 x64
• Python 3.7.3 (x64)
• TensorFlow-GPU 1.13.1 ([TensorFlow]: Install TensorFlow with pip)

Error:

[cfati@CFATI-5510-0:e:\Work\Dev\StackOverflow\q057588589]> "e:\Work\Dev\VEnvs\py_064_03.07.03_test0\Scripts\python.exe" -c "import tensorflow as tf"
Traceback (most recent call last):
  File "e:\Work\Dev\VEnvs\py_064_03.07.03_test0\lib\site-packages\tensorflow\python\pywrap_tensorflow.py", line 58, in <module>
    from tensorflow.python.pywrap_tensorflow_internal import *
  File "e:\Work\Dev\VEnvs\py_064_03.07.03_test0\lib\site-packages\tensorflow\python\pywrap_tensorflow_internal.py", line 28, in <module>
    _pywrap_tensorflow_internal = swig_import_helper()
  File "e:\Work\Dev\VEnvs\py_064_03.07.03_test0\lib\site-packages\tensorflow\python\pywrap_tensorflow_internal.py", line 24, in swig_import_helper
    _mod = imp.load_module('_pywrap_tensorflow_internal', fp, pathname, description)
  File "e:\Work\Dev\VEnvs\py_064_03.07.03_test0\lib\imp.py", line 242, in load_module
    return load_dynamic(name, filename, file)
  File "e:\Work\Dev\VEnvs\py_064_03.07.03_test0\lib\imp.py", line 342, in load_dynamic
    return _load(spec)
ImportError: DLL load failed: The specified module could not be found.

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "e:\Work\Dev\VEnvs\py_064_03.07.03_test0\lib\site-packages\tensorflow\__init__.py", line 24, in <module>
    from tensorflow.python import pywrap_tensorflow  # pylint: disable=unused-import
  File "e:\Work\Dev\VEnvs\py_064_03.07.03_test0\lib\site-packages\tensorflow\python\__init__.py", line 49, in <module>
    from tensorflow.python import pywrap_tensorflow
  File "e:\Work\Dev\VEnvs\py_064_03.07.03_test0\lib\site-packages\tensorflow\python\pywrap_tensorflow.py", line 74, in <module>
    raise ImportError(msg)
ImportError: Traceback (most recent call last):
  File "e:\Work\Dev\VEnvs\py_064_03.07.03_test0\lib\site-packages\tensorflow\python\pywrap_tensorflow.py", line 58, in <module>
    from tensorflow.python.pywrap_tensorflow_internal import *
  File "e:\Work\Dev\VEnvs\py_064_03.07.03_test0\lib\site-packages\tensorflow\python\pywrap_tensorflow_internal.py", line 28, in <module>
    _pywrap_tensorflow_internal = swig_import_helper()
  File "e:\Work\Dev\VEnvs\py_064_03.07.03_test0\lib\site-packages\tensorflow\python\pywrap_tensorflow_internal.py", line 24, in swig_import_helper
    _mod = imp.load_module('_pywrap_tensorflow_internal', fp, pathname, description)
  File "e:\Work\Dev\VEnvs\py_064_03.07.03_test0\lib\imp.py", line 242, in load_module
    return load_dynamic(name, filename, file)
  File "e:\Work\Dev\VEnvs\py_064_03.07.03_test0\lib\imp.py", line 342, in load_dynamic
    return _load(spec)
ImportError: DLL load failed: The specified module could not be found.


Failed to load the native TensorFlow runtime.

See https://www.tensorflow.org/install/errors

for some common reasons and solutions.  Include the entire stack trace
above this error message when asking for help.

Looking at the "faulty" module (thanks to Dependency Walker), it turns out that it's not itself that's missing, but some of its dependencies (the cu*_100.dll files).

Check [SO]: Python Ctypes - loading dll throws OSError: [WinError 193] %1 is not a valid Win32 application (@CristiFati's answer) (the Conclusions section at the end) for more details on this kind of errors.
I had an older CUDA Toolkit version (8), and as a consequence, the cu*_80.dll files.

Upgrading to TensorFlow-GPU 1.14.0 ("e:\Work\Dev\VEnvs\py_064_03.07.03_test0\Scripts\python.exe" -m pip install --upgrade tensorflow-gpu), made the error a bit clearer (and also shorter):

[cfati@CFATI-5510-0:e:\Work\Dev\StackOverflow\q057588589]> "e:\Work\Dev\VEnvs\py_064_03.07.03_test0\Scripts\python.exe" -c "import tensorflow as tf"
Traceback (most recent call last):
  File "e:\Work\Dev\VEnvs\py_064_03.07.03_test0\lib\site-packages\tensorflow\python\platform\self_check.py", line 75, in preload_check
    ctypes.WinDLL(build_info.cudart_dll_name)
  File "c:\install\x64\python\python\03.07.03\Lib\ctypes\__init__.py", line 356, in __init__
    self._handle = _dlopen(self._name, mode)
OSError: [WinError 126] The specified module could not be found

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "e:\Work\Dev\VEnvs\py_064_03.07.03_test0\lib\site-packages\tensorflow\__init__.py", line 28, in <module>
    from tensorflow.python import pywrap_tensorflow  # pylint: disable=unused-import
  File "e:\Work\Dev\VEnvs\py_064_03.07.03_test0\lib\site-packages\tensorflow\python\__init__.py", line 49, in <module>
    from tensorflow.python import pywrap_tensorflow
  File "e:\Work\Dev\VEnvs\py_064_03.07.03_test0\lib\site-packages\tensorflow\python\pywrap_tensorflow.py", line 30, in <module>
    self_check.preload_check()
  File "e:\Work\Dev\VEnvs\py_064_03.07.03_test0\lib\site-packages\tensorflow\python\platform\self_check.py", line 82, in preload_check
    % (build_info.cudart_dll_name, build_info.cuda_version_number))
ImportError: Could not find 'cudart64_100.dll'. TensorFlow requires that this DLL be installed in a directory that is named in your %PATH% environment variable. Download and install CUDA 10.0 from this URL: https://developer.nvidia.com/cuda-90-download-archive

Steps:

• Uninstall any CUDA Toolkit version (optional)
• Install [nVidia.Developer]: CUDA Toolkit 10.0 Archive
  • Make sure to install v10.0 (that this TensorFlow-GPU version was built against). I installed v10.1 (which was the latest, and also the recommended version at the answer time), and the .dll names didn't match (cu*_101.dll). Since I didn't want to install v10.0, I created some symlinks (with the "correct" names) to the existing files, and it worked. But bear in mind that this is unsupported!!! You may experience funny behavior (including crashes). This is a (lame) workaround (gainarie)
  • Additionally, a compatible (meaning that it's for a specific CUDA Toolkit version) cuDNN version ([nVidia.Developer]: cuDNN Archive) is required. In order to access the download URL, nVidia membership is required

After the above steps, and also setting the correct paths, it worked:

[cfati@CFATI-5510-0:e:\Work\Dev\StackOverflow\q057588589]> set PATH=%PATH%;%CUDA_PATH%\bin;f:\Install\x64\NVidia\GPU Computing Toolkit\cuDNN\7.6\bin

[cfati@CFATI-5510-0:e:\Work\Dev\StackOverflow\q057588589]> "e:\Work\Dev\VEnvs\py_064_03.07.03_test0\Scripts\python.exe" -c "import tensorflow;print(\"Success!!!\")"
Success!!!

cuDNN causes my problem. PATH variable doesn't work for me. I have to copy the files in my cuDNN folders into respectful CUDA 8.0 folder structure.

For Those Running on Older Hardware:

You may get this same error due to having an older CPU using tensorflow-gpu 1.6.

If your cpu was made before 2011, then your max tensorflow-gpu version is 1.5.

Tensorflow 1.6 requires AVX instructions on your cpu. Verified here: Tensorflow Github docs

AVX enabled CPUs: Wiki AVX CPUs

What I did in my conda environment for tensorflow:

pip install --ignore-installed --upgrade tensorflow-gpu==1.5

For tensorflow with CPU only:

I had installed tensorflow using command:

pip3 install --upgrade tensorflow

This installed tensorflow 1.7
But could not import the tensorflow from withing python 3.6.5 amd64 using:

import tensorflow as tf

So, i downgraded the tensorflow version from 1.7 to 1.5 using following command:

pip3 install tensorflow==1.5

This uninstalled the previous version and installed 1.5. Now it works.

Seems that, my CPU does not support AVX instruction set that is needed in tensorflow 1.7

I had MSVCP140.DLL in the system folders and .DLL in the PATHEXT variable in Environment Variable.

The problem was the cuDNN Library for me. I was able to run the test code after adding the directory (possibly bin folder) of the cuDNN DLL (not LIB file) in the Windows PATH.

For the reference, I installed TensorFlow from the source using PIP and my OS: Windows 7 and IDE: Visual Studio 2015.

With TensorFlow release 1.3.0, you need to use Cudnn 6.0 instead of Cudnn 5.0 as Cudnn 5.0 is giving this error. Don't forget to add path variable to Cudnn 6.0 .With cudnn64_6.dll your Tensorflow will work fine. Read the link below. https://github.com/tensorflow/tensorflow/blob/master/RELEASE.md#release-130

my answer is for windows 10 users only as I have tried the following on windows 10. Extending some of the answers above I suggest this : If you are using anaconda then you can avoid everything and simply install anaconda-navigator using the command

conda install -c anaconda anaconda-navigator

Then you can launch the navigator from command prompt using the command

anaconda-navigator

On running this command you get a simple gui where you can create an virtual environment, create the environment with python=3.5.2 and install module tensorflow-gpu or tensorflow by searching the module in the search box using gui, it will also take care of installing correct cuda files for you. Using anaconda navigator is the simplest solution.

If you are not using anaconda then take care about the following

tensorflow-gpu 1.3 requires python 3.5.2, cuda development kit 8.0 and cudaDNN 6.0, hence when installing make sure you run the command

pip install tensorflow-gpu==1.3

tensorflow-gpu 1.2.1 or less requires python 3.5.2, cuda development kit 8.0 and cudaDNN 5.1 hence when installing make sure you run the command

pip install tensorflow-gpu==1.2.1

Below are the steps you need to follow for both of the above processes Setting up you path variables You must have the following system variables

CUDA_HOME = "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0"
CUDA_PATH = "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0"
CUDA_PATH_V8.0 = "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0"

You PATHTEXT must include ".DLL" along with other extensions

".COM;.EXE;.BAT;.CMD;.VBS;.VBE;.JS;.JSE;.WSF;.WSH;.MSC;.PY;.DLL"

Also Add the following to you path

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\lib\x64
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\extras\CUPTI\libx64;
C:\Windows\SysWOW64;
C:\Windows\System32        

If you are getting errors you can download the run the below code by mrry, this code will check your setup and tell you if something is wrong https://gist.github.com/mrry/ee5dbcfdd045fa48a27d56664411d41c

References : http://blog.nitishmutha.com/tensorflow/2017/01/22/TensorFlow-with-gpu-for-windows.html

The above reference is very useful. Please comment for improvements to this answer. Hope this helps, Thanks.

The problem was the cuDNN Library for me - for whatever reason cudnn-8.0-windows10-x64-v6.0 was NOT working - I used cudnn-8.0-windows10-x64-v5.1 - ALL GOOD!

My setup working with Win10 64 and the Nvidia GTX780M:

• Be sure you have the lib MSVCP140.DLL by checking your system/path - if not get it here
• Run the windows installer for python 3.5.3-amd64 from here - DO NOT try newer versions as they probably won't work
• Get the cuDNN v5.1 for CUDA 8.0 from here - put it under your users folder or in another known location (you will need this in your path)
• Get CUDA 8.0 x86_64 from here
• Set PATH vars as expected to point at the cuDNN libs and python (the python path should be added during the python install)
• Make sure that ".DLL" is included in your PATHEXT variable
• If you are using tensorflow 1.3 then you want to use cudnn64_6.dll github.com/tensorflow/tensorflow/issues/7705

If you run Windows 32 be sure to get the 32 bit versions of the files mentioned above.

One may be tempted to keep the Powershell/cmd open on Windows. I've spent reasonable time till I decided to close and reopen my Powershell only to realize that I've done everything right.

For the people finding this post in 2019, this error could also occur because the Python version 3.7 does not have support for TensorFlow (see https://www.tensorflow.org/install/pip). So, check the Python version:

python --version

In case it is larger than 3.6, it should be downgraded to 3.6. For Anaconda:

conda install python=3.6

Then, install TensorFlow.

pip install tensorflow

Btw, I did not have the GPU version, so there were no CUDA related issues in my case.

For each Tensorflow's version, it requires different version of CuDnn. On www.tensorflow.org, they did not mentioned about that in installation guide!

My case use tensorflow version 1.3 which uses cuDNN 6. https://github.com/tensorflow/tensorflow/releases.

Please check your tensorfow version and cuDNN version if they are match together.

And please set path environment for cuDNN, if it still does not work, please check the answer from @Chris Han.

Either error indicates that your system has not installed MSVCP140.DLL, which TensorFlow requires.

To fix this error:

1. Determine whether MSVCP140.DLL is in your %PATH% variable.
2. If MSVCP140.DLL is not in your %PATH%, install the Visual C++ 2015 redistributable (x64 version), which contains this DLL.