Creating a specific 3.6 binary for Linux

[KrishnaveniTK]

System information

• Have I written custom code (as opposed to using a stock example script provided in TensorFlow): No
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Linux Ubuntu 14.04
• TensorFlow installed from (source or binary): Source
• TensorFlow version (use command below):
• Python version: Python3.6
• Bazel version (if compiling from source):
• GCC/Compiler version (if compiling from source): -NA-
• CUDA/cuDNN version: -NA-
• GPU model and memory: -NA-
• Exact command to reproduce: import tensorflow as tf

Environment capture text:

https://github.com/tensorflow/tensorflow/tree/master/tools/tf_env_collect.sh
tf_env.txt

You can obtain the TensorFlow version with
python -c "import tensorflow as tf; print(tf.GIT_VERSION, tf.VERSION)"
This command also results in the same error.
/home/raju/anaconda3/lib/python3.6/importlib/_bootstrap.py:219: RuntimeWarning: compiletime version 3.5 of module 'tensorflow.python.framework.fast_tensor_util' does not match runtime version 3.6
return f(*args, **kwds)

Describe the problem

When importing tensorflow , I get this error. I found some information on "Feature request: nightly build for python 3.6 #12935" --"Yes, we unfortunately copy the 3.5 binary for 3.6 I'll look into creating a specific 3.6 binary for Linux."

Source code / logs

$import tensorflow as tf
result is
/home/raju/anaconda3/lib/python3.6/importlib/_bootstrap.py:219: RuntimeWarning: compiletime version 3.5 of module 'tensorflow.python.framework.fast_tensor_util' does not match runtime version 3.6
return f(*args, **kwds)

[jbboisseau]

same issue on MacOS 10.12.6

[rfarouni]

same problem on Linux Ubuntu 16.04

[asimshankar]

@gunan : Do we have any updates on the resources we have to support Python 3.6 releases?

[gunan]

@av8ramit looks like there were divergences which prevent us from using the same pip packages on python 3.5 and 3.6
Can we modify our builds to also build 3.6 version from scratch?

[YannDubs]

Same issue on Linux CentOS 7

[knn1989]

Same issue after upgrade Tensorflow from 1.3 to 1.4 on Mac High Sierra 10.13.1 using Python 3.6.2.
Tensorflow still works but give the above warning which is "RuntimeWarning: compiletime version 3.5 of module 'tensorflow.python.framework.fast_tensor_util' does not match runtime version 3.6"

[MichaelSnowden]

FWIW, python3.5 still works

[sbshah97]

ANything that can be changed here?

[wieberinsma]

Same problem using virtualenv after upgrading from Ubuntu 17.04 to 17.10.
/home/wiebe/virtual/tensorflow/lib/python3.6/importlib/_bootstrap.py:219: RuntimeWarning: compiletime version 3.5 of module 'tensorflow.python.framework.fast_tensor_util' does not match runtime version 3.6 return f(*args, **kwds)

[banifou]

Same on Ubuntu 14.04

[dsitnik]

Same issue on Fedora 26 with Python 3.6.2.

[seanxwzhang]

Same issue on OSX 10.12.5 with Python 3.6.3

[nqiao]

Same issue on OSX with Python 3.6.2

[erichiller]

Only posting because I didn't see my environment mentioned:
Clean 17.10 Ubuntu Server install on Python 3.6.3 (native pip install tensorflow-gpu)
Outputs:

/usr/lib/python3.6/importlib/_bootstrap.py:219: RuntimeWarning: compiletime version 3.5 of module 'tensorflow.python.framework.fast_tensor_util' does not match runtime version 3.6
  return f(*args, **kwds)

[h-jia]

I have tried both 3.5 and 3.6 (macOS High Sierra).
3.5 works perfect while 3.6 gives warns.
While 3.6 still works.

Hope this helps.

[AlphaSmartDog]

same problem on Linux Ubuntu 17.10

Clean 17.10 Ubuntu Server install on Python 3.6.3 (native pip install tensorflow)

[february]

Same issue on OSX 10.12.6 with Python 3.6.3(virtualenv installation)

import tensorflow as tf
/Users/zhanghao/virtualenv/tensorflow/lib/python3.6/importlib/_bootstrap.py:219: RuntimeWarning: compiletime version 3.5 of module 'tensorflow.python.framework.fast_tensor_util' does not match runtime version 3.6
return f(*args, **kwds)

[christiangda]

Same issue using this Docker file https://github.com/christiangda/kerasvideo/blob/master/Dockerfile

error:

import tensorflow as tf

/opt/conda/lib/python3.6/importlib/_bootstrap.py:219: RuntimeWarning: compiletime version 3.5 of module 'tensorflow.python.framework.fast_tensor_util' does not match runtime version 3.6
  return f(*args, **kwds)

[KrishnaveniTK]

Hi,
Can you let me know if this has to be treated as a warning and ignored?
Furthermore the error is
from PyQt5 import QtCore, QtGui, QtWidgets
ImportError: dlopen: cannot load any more object with static TLS

See the Traceback below
Traceback (most recent call last):
File "train.py", line 51, in
from object_detection.builders import model_builder
File "/home/raju/anaconda3/envs/tensorflow/models/research/object_detection/builders/model_builder.py", line 29, in
from object_detection.meta_architectures import ssd_meta_arch
File "/home/raju/anaconda3/envs/tensorflow/models/research/object_detection/meta_architectures/ssd_meta_arch.py", line 31, in
from object_detection.utils import visualization_utils
File "/home/raju/anaconda3/envs/tensorflow/models/research/object_detection/utils/visualization_utils.py", line 24, in
import matplotlib.pyplot as plt
File "/home/raju/anaconda3/lib/python3.6/site-packages/matplotlib/pyplot.py", line 113, in
_backend_mod, new_figure_manager, draw_if_interactive, _show = pylab_setup()
File "/home/raju/anaconda3/lib/python3.6/site-packages/matplotlib/backends/init.py", line 60, in pylab_setup
[backend_name], 0)
File "/home/raju/anaconda3/lib/python3.6/site-packages/matplotlib/backends/backend_qt5agg.py", line 16, in
from .backend_qt5 import (
File "/home/raju/anaconda3/lib/python3.6/site-packages/matplotlib/backends/backend_qt5.py", line 18, in
import matplotlib.backends.qt_editor.figureoptions as figureoptions
File "/home/raju/anaconda3/lib/python3.6/site-packages/matplotlib/backends/qt_editor/figureoptions.py", line 20, in
import matplotlib.backends.qt_editor.formlayout as formlayout
File "/home/raju/anaconda3/lib/python3.6/site-packages/matplotlib/backends/qt_editor/formlayout.py", line 56, in
from matplotlib.backends.qt_compat import QtGui, QtWidgets, QtCore
File "/home/raju/anaconda3/lib/python3.6/site-packages/matplotlib/backends/qt_compat.py", line 137, in
from PyQt5 import QtCore, QtGui, QtWidgets
ImportError: dlopen: cannot load any more object with static TLS

Pls let me know ways to resolve this error. Or will it be resolved only when the warning is resolved!!

[laoshaw]

Same issues on Ubuntu 16.04 64bit virtualenv python3

[brian8128]

Does it break stuff or can I ignore the warning?

[dartdog]

Quite strange since the prior version worked fine with 3.6 (and it's packages), so it would seem that this should be a pretty simple fix?

[laoshaw]

same issue with the just released 1.4 version...

[9034725985]

Just to add to what @dsitnik said, simple hello world to reproduce on Fedora 26 venv

$ cat requirements.txt 
bleach==1.5.0
enum34==1.1.6
html5lib==0.9999999
Markdown==2.6.9
numpy==1.13.3
protobuf==3.4.0
six==1.11.0
tensorflow==1.4.0
tensorflow-tensorboard==0.4.0rc2
Werkzeug==0.12.2

$ python --version
Python 3.6.2

$ cat app.py 
import tensorflow as tf

hello = tf.constant('Hello, TensorFlow!')
session = tf.Session()
print(session.run(hello))

$ python app.py 
/usr/lib64/python3.6/importlib/_bootstrap.py:205: RuntimeWarning: compiletime version 3.5 of module 'tensorflow.python.framework.fast_tensor_util' does not match runtime version 3.6
  return f(*args, **kwds)
2017-11-07 16:41:00.517173: I tensorflow/core/platform/cpu_feature_guard.cc:137] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2
b'Hello, TensorFlow!'

[GreggHelt2]

Same issue on OSX 10.11.6, in an Anaconda virtual environment with Python 3.6.3 and attempting upgrade from TensorFlow 1.3 to 1.4.
Temporarily routed around this issue by creating a new Anaconda environment with Python 3.5.4 and Tensorflow 1.4, which is working fine.

[auroua]

Same issue.

[quoniammm]

Same issue on Ubuntu 16.04 LTS.

[av8ramit]

We are not offering a native python3.6 binary for mac anytime soon.

As for the cuda9 issue, tf-nightly is built from HEAD at master every night, and I believe the latest binaries require cuda9. Try using an older version of tf-nightly. (Maybe 1208 or something)

Windows has a native python3.6 binary so you will not see this issue.

[tylerlekang]

Thanks @av8ramit

For those who have the (arbitrary) preference of only wanting to use official releases, the following is correct?

• you'll need to wait for 1.5.x for this warning message to go away when using Python 3.6 on linux
• in the meantime, you can use 1.4.1 with no problems, you'll just see the warning message

[aishoot]

The same issue appears while I use Anaconda 3-5.0.1-Linux-x86_64.sh(Python3.6) and TensorFlow 1.4.

[eric-nok]

For what it's worth, the issue was discussed in a Hacker News thread about the 1.4 release (https://news.ycombinator.com/item?id=15647790). User allenlavoie responded:

It's just a small Cython-compiled utility throwing the warning, it works fine. Everything else is pure Python and so doesn't care about 3.5 vs 3.6. We will fix the warning, though (disclaimer: I work on TF and added fast_tensor_utils--sorry!).

[nji-syd]

Dear Tensorflow QA team. Please add some unit tests!!! Happy NY

[arunmandal53]

You can create whl from source using following link steps. http://www.python36.com/install-tensorflow141-gpu/ . Let me informed if it worked.

[tensorflowbutler]

Nagging Assignee: It has been 14 days with no activity and this issue has an assignee. Please update the label and/or status accordingly.

[av8ramit]

Official releases now (1.5+) have a native python3.6 image build on Ubuntu 16.04. We do not build a python3.6 specific binary for mac. Windows has native 3.6 as well.

[tylerlekang]

@av8ramit , I have 1.4.1 and you said that official releases now have .... yet,

When I do:
python3.6 -m pip install --upgrade --user tensorflow

It says that:
Requirement already up-to-date: tensorflow in ./.local/lib/python3.6/site-packages

[stevenlii]

You guys could get python3.6 specific binary for mac/linux/windows from here:
https://pypi.python.org/pypi/tensorflow

[tylerlekang]

@av8ramit I manually downloaded 1.5.0rc1 cp36 manylinux wheel and installed successfully. The warning message has disappeared when import tensorflow in python 3.6

However, I think there is issue with pip. See below please. Do you need me to start a new issue, or am I just doing something wrong?

When I do:
python3.6 -m pip install --upgrade --user tensorflow -vvv

It does find the 1.5.0rc1 cp36 manylinux wheel, as well as rc0 and older version wheels (both rc and non-rc):

Found link https://pypi.python.org/packages/cd/e4/b2a8bcd1fa689489050386ec70c5c547e4a75d06f2cc2b55f45463cd092c/tensorflow-1.1.0-cp36-cp36m-manylinux1_x86_64.whl#md5=1f761290358dfb7fe4ec73140f4d282a (from https://pypi.python.org/simple/tensorflow/), version: 1.1.0
...
Found link https://pypi.python.org/packages/c6/1c/d64f9367cc3a50194e911ab9da865c2515cbc1fefa5e210ddebf85a61bac/tensorflow-1.2.0rc0-cp36-cp36m-manylinux1_x86_64.whl#md5=8a637f414bc16c86708c01f17c94f3f6 (from https://pypi.python.org/simple/tensorflow/), version: 1.2.0rc0
...
Found link https://pypi.python.org/packages/fe/7d/01c588ba060e5e7dffe64ab968745d143530c84da30296f9593f661ca27d/tensorflow-1.5.0rc0-cp36-cp36m-manylinux1_x86_64.whl#md5=6578302e0c843660464cc16ad45d6093 (from https://pypi.python.org/simple/tensorflow/), version: 1.5.0rc0
...
Found link https://pypi.python.org/packages/4e/50/87e3c6861957413deb5ccd62aee6cea85cc2eedcf6391183595191d3c87d/tensorflow-1.5.0rc1-cp36-cp36m-manylinux1_x86_64.whl#md5=501d5072275518056aa5a5f25a001015 (from https://pypi.python.org/simple/tensorflow/), version: 1.5.0rc1

but at the end this does not register somehow with pip to be the latest version:

Installed version (1.4.1) is most up-to-date (past versions: 0.12.1, 1.0.0, 1.0.1, 1.1.0, 1.2.0, 1.2.1, 1.3.0, 1.4.0, 1.4.1)
Requirement already up-to-date: tensorflow in ./.local/lib/python3.6/site-packages

I may be completely wrong, but it seems pip is not catching the versions with rc in the version number. I don't know if this is pip's error or tensorflow's error, but it seems to be an issue. Or am I wrong?

[attiasr]

@tylerlekang pip fetch the latest stable release by default (see https://pip.pypa.io/en/stable/reference/pip_install/#pre-release-versions)
and the pypi package of tensorflow is still a release candidate (1.5.0rc1)

try using:
pip install 'tensorflow>=1.5.0rc1'

[tylerlekang]

@attiasr Oh! Well ... that's embarrassing!

How long does it take the rc to turn into stable, and are there usually any changes once released as rc? Well I guess I can answer my own question by fact that there has already been rc0 and rc1 of 1.5.0 .....

Thanks!

[attiasr]

@tylerlekang np

[molyswu]

python3.6 -m pip install https://github.com/lakshayg/tensorflow-build/raw/master/tensorflow-1.4.0-cp36-cp36m-linux_x86_64.whl

[ozkalt]

Thanks @Karthick333031 it works for me (MacOS 10.12.6)

[umang6891]

Running: pip3 install --upgrade https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-1.10.0-py3-none-any.whl solved the issue

Python 3.6.5
TensorFlow 1.10.0 (Was previously 1.10.1)
MacOS 10.14

[hectorip]

Thanks,

@umang6891! It solved the problem for me. With the warning, my example hanged forever without failing, but now it runs. It looks like the problem is the version in pip.

👍

[zhangjunpeng9354]

@umang6891 's solution works good for me, and I am using a python3.6 virtualenv.