Bug in tensorflow tuttorial about quantization

Question

This post follow my another post about op not registered: : NotFoundError: Op type not registered 'Dequantize'

And now when I try load manually as in this post: Install Tensorflow with Quantization Support

with:

qops = tf.load_op_library('/home/artem/projects/tensorflow/bazel-bin/tensorflow/contrib/quantization/_quantized_ops.so')
qkernelops = tf.load_op_library('/home/artem/projects/tensorflow/bazel-bin/tensorflow/contrib/quantization/kernels/_quantized_kernels.so')

I get the following error:

---------------------------------------------------------------------------
NotFoundError                             Traceback (most recent call last)
<ipython-input-4-799f99708cda> in <module>()
      1 
----> 2 qops = tf.load_op_library('/home/artem/projects/tensorflow/bazel-bin/tensorflow/contrib/quantization/_quantized_ops.so')
      3 qkernelops = tf.load_op_library('/home/artem/projects/tensorflow/bazel-bin/tensorflow/contrib/quantization/kernels/_quantized_kernels.so')

/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/load_library.pyc in load_op_library(library_filename)
     69           return _OP_LIBRARY_MAP[library_filename]
     70       # pylint: disable=protected-access
---> 71       raise errors._make_specific_exception(None, None, error_msg, error_code)
     72       # pylint: enable=protected-access
     73   finally:

NotFoundError: /home/artem/projects/tensorflow/bazel-bin/tensorflow/contrib/quantization/_quantized_ops.so: undefined symbol: _ZN10tensorflow8internal21CheckOpMessageBuilder9NewStringB5cxx11Ev

It is look like that it is a problem with C++ compiler and probably I should add -D_GLIBCXX_USE_CXX11_ABI=0 as in this post: https://github.com/tensorflow/tensorflow/pull/1584

but actually I did bazel-build and I don't know how manage it.

Anybody can help me? Test example from https://www.tensorflow.org/versions/master/how_tos/quantization/index.html also don't work. It will be great if someone would write detailed manual about quantization generally.

sounds like a good subject for github issue, mentioning @petewarden — Yaroslav Bulatov, Jul 27 '16 at 17:11
Thank you for advice. I create it here: https://github.com/tensorflow/tensorflow/issues/3533 — Artem Grachev, Jul 27 '16 at 17:55

score 2 · Accepted Answer · answered Aug 02 '16 at 19:38

Tensorflow 0.10 is released! I solved this problem after updating! Add those code in your script:

from tensorflow.contrib.quantization import load_quantized_ops_so
from tensorflow.contrib.quantization.kernels import load_quantized_kernels_so

and

load_quantized_ops_so.Load()
load_quantized_kernels_so.Load()

Thanks the TF teams!

Bug in tensorflow tuttorial about quantization

1 Answers1

Linked