TensorFlow 1.7 whl issue on TX2

Robotics & Edge Computing Jetson & Embedded Systems Jetson TX2
1

Hi
I build Tensorflow 1.7 on my TX2. i installed it using sudo pip3 install tensorflow-1.7.0-cp35-cp35m-linux_aarch64.whl. THere were no errors as you see below

Successfully installed tensorboard-1.7.0 tensorflow-1.7.0

In PYthon3 when i say import tensorflow as tf I am getting below error

Traceback (most recent call last):
File “/usr/local/lib/python3.5/dist-packages/tensorflow/python/pywrap_tensorflow.py”, line 58, in
from tensorflow.python.pywrap_tensorflow_internal import *
File “/usr/local/lib/python3.5/dist-packages/tensorflow/python/pywrap_tensorflow_internal.py”, line 28, in
_pywrap_tensorflow_internal = swig_import_helper()
File “/usr/local/lib/python3.5/dist-packages/tensorflow/python/pywrap_tensorflow_internal.py”, line 24, in swig_import_helper
_mod = imp.load_module(‘_pywrap_tensorflow_internal’, fp, pathname, description)
File “/usr/lib/python3.5/imp.py”, line 242, in load_module
return load_dynamic(name, filename, file)
File “/usr/lib/python3.5/imp.py”, line 342, in load_dynamic
return _load(spec)
ImportError: /usr/local/lib/python3.5/dist-packages/tensorflow/python/_pywrap_tensorflow_internal.so: undefined symbol: _ZN3Aws11Environment6GetEnvB5cxx11EPKc

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File “”, line 1, in
File “/usr/local/lib/python3.5/dist-packages/tensorflow/init.py”, line 24, in
from tensorflow.python import * # pylint: disable=redefined-builtin
File “/usr/local/lib/python3.5/dist-packages/tensorflow/python/init.py”, line 49, in
from tensorflow.python import pywrap_tensorflow
File “/usr/local/lib/python3.5/dist-packages/tensorflow/python/pywrap_tensorflow.py”, line 74, in
raise ImportError(msg)
ImportError: Traceback (most recent call last):
File “/usr/local/lib/python3.5/dist-packages/tensorflow/python/pywrap_tensorflow.py”, line 58, in
from tensorflow.python.pywrap_tensorflow_internal import *
File “/usr/local/lib/python3.5/dist-packages/tensorflow/python/pywrap_tensorflow_internal.py”, line 28, in
_pywrap_tensorflow_internal = swig_import_helper()
File “/usr/local/lib/python3.5/dist-packages/tensorflow/python/pywrap_tensorflow_internal.py”, line 24, in swig_import_helper
_mod = imp.load_module(‘_pywrap_tensorflow_internal’, fp, pathname, description)
File “/usr/lib/python3.5/imp.py”, line 242, in load_module
return load_dynamic(name, filename, file)
File “/usr/lib/python3.5/imp.py”, line 342, in load_dynamic
return _load(spec)
ImportError: /usr/local/lib/python3.5/dist-packages/tensorflow/python/_pywrap_tensorflow_internal.so: undefined symbol: _ZN3Aws11Environment6GetEnvB5cxx11EPKc

i looked upt Build from source  |  TensorFlow but it is of no help.

I searched google but still of no help

I am running CUDA 9.0 Cudnn 7.0.5 and PYthon 3.5

I tried with Tensorflow 1.6 and also 1.5 but same issue

Hoping someone can help me here. Thanks in advance

2

You may look through forum threads on tensorflow installation.
I am inclined to believe there were whl shared at a number of threads.
You may use for reference:
https://devtalk.nvidia.com/default/topic/1031300/tensorflow-1-7-wheel-with-jetpack-3-2-/
You may also try install 27 python tensorflow.
As far as I am concerned you may need to export PATH’s of either python or CUDA or both.
somewhat github: repo GitHub - peterlee0127/tensorflow-nvJetson: TensorFlow for NVIDIA Jetson, also include patch and script for building.

3

well i downloaded whl file from https://devtalk.nvidia.com/default/topic/1031300/jetson-tx2/tensorflow-1-7-wheel-with-jetpack-3-2-/?offset=29#5254966

i uninstalled mine and installed the one i downloaded. Worke good with python3.5. Thanks though for help

4

Same error as above

_pywrap_tensorflow_internal.so: undefined symbol: _ZN3Aws11Environment6GetEnvB5cxx11EPKc

Setup: ubuntu 16.04, Jetson TX1, Python3.5, Bazel 0.17.2

git clone GitHub - mozilla/tensorflow: Computation using data flow graphs for scalable machine learning
./configure (only enabled CUDA)
bazel build --config=opt --config=cuda //tensorflow/tools/pip_package:build_pip_package

After several hours, get the same import error as above with AWS. As per this link Build Tensorflow v1.7 on NVIDIA Jetson tx2, they suggest disabling S3 but I never enabled it when running ./configure.

I can’t use py 2.7 and need to build tensorflow source.

Thanks for your time.

5

If you are not going to use AWS stuffs with tensorflow, you could try saying no to the following question during ‘./configure’. See if that solves the problem.

Do you wish to build TensorFlow with Amazon S3 File System support? [Y/n]: n

Here’s how I built tensorflow-1.8.0 for python3 on JTX2. FYR.

https://jkjung-avt.github.io/build-tensorflow-1.8.0/

6

Thanks @ jkjung13

Definitely not using AWS. But with the tensorflow version that am using (git clone GitHub - mozilla/tensorflow: Computation using data flow graphs for scalable machine learning), I don’t even get that S3 option during the configure step.

Let me uninstall and try your process with the mozilla tensorflow (v 1.11). Since the S3/AWS option doesn’t appear, am editing .tf_configure.bazelrc to explicitly disable S3

build --action_env TF_NEED_S3=“0”

For some odd reason that still tries to compile aws but not when I export the variable.

export TF_NEED_S3=0

I’ll share the results.

7

With the latest tensorflow from Mozilla, first got the error:

gcc: error: unrecognized command line option ‘-mfpu=neon’

so got rid of that option in the file /tensorflow/contrib/lite/kernels/internal/BUILD

and 4-5 hrs later got the same error as before

ImportError: /home/nvidia/.cache/bazel/_bazel_nvidia/d2751a49dacf4cb14a513ec663770624/execroot/org_tensorflow/bazel-out/host/bin/tensorflow/create_tensorflow.python_api_1.runfiles/org_tensorflow/tensorflow/python/_pywrap_tensorflow_internal.so: undefined symbol: _ZN3Aws11Environment6GetEnvB5cxx11EPKc

Failed to load the native TensorFlow runtime.

See Build and install error messages  |  TensorFlow

for some common reasons and solutions. Include the entire stack trace
above this error message when asking for help.
Target //tensorflow/tools/pip_package:build_pip_package failed to build

And just a few lines before the above error, get this:

ERROR: /home/nvidia/tensorflow/tensorflow/BUILD:557:1: Executing genrule //tensorflow:tf_python_api_gen_v1 failed (Exit 1)
Traceback (most recent call last):
File “/home/nvidia/.cache/bazel/_bazel_nvidia/d2751a49dacf4cb14a513ec663770624/execroot/org_tensorflow/bazel-out/host/bin/tensorflow/create_tensorflow.python_api_1.runfiles/org_tensorflow/tensorflow/python/pywrap_tensorflow.py”, line 58, in
from tensorflow.python.pywrap_tensorflow_internal import *

So disabling S3 via the env. variable and setting TF_NEED_S3=0 does not appear to suffice to prevent building the S3 filesystem.

Still debugging …

8

Hi,

Not sure the difference between official and Mozilla version.

Here is a script to build official v1.6 from source for your reference:
https://github.com/AastaNV/JEP/tree/master/script/TensorFlow_1.6

Thanks.

9

Thanks. Actually got the latest tensorflow version (1.11) to build successfully with the following steps:

In the BUILD file here: /home/<user_name>/.cache/bazel/bazel<user_name>//external/aws/BUILD.bazel
if I edit the section cc_library to replace:

“//conditions:default”: "

with

“//conditions:default”: glob([“aws-cpp-sdk-core/source/platform/linux-shared/*.cpp”,]),

So there appears to be build issues with the AWS SDK on aarch64. I followed instructions from these two links:

https://devtalk.nvidia.com/default/topic/1031642/jetson-tx2/tensorflow-serving-on-jetson-tx2/
https://github.com/tensorflow/serving/issues/832

Let me see if there are any runtime issues and if there are, then I’ll try tf v1.6 as per instructions in the AastaNV link.