frame

Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

Sign In

Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

Incompatible Tensorflow versions

What version of Tensorflow and Python are being installed with the MNC SDK 2.04?

I'm having issues that seem to come from incompatibility between the TF version for training and inference on the MNCS. The error seem to suggest that inference uses python 3.5.

Training is done on TF 1.6 and python 2.7

/usr/local/lib/python3.5/dist-packages/tensorflow/python/framework/ops.py:766: DeprecationWarning: builtin type EagerTensor has no __module__ attribute
  EagerTensor = c_api.TFE_Py_InitEagerTensor(_EagerTensorBase)
/usr/local/lib/python3.5/dist-packages/tensorflow/python/util/tf_inspect.py:45: DeprecationWarning: inspect.getargspec() is deprecated, use inspect.signature() instead
  if d.decorator_argspec is not None), _inspect.getargspec(target))
2018-06-08 22:06:46.052137: E tensorflow/core/common_runtime/executor.cc:643] Executor failed to create kernel. Invalid argument: NodeDef mentions attr 'dilations' not in Op<name=Conv2D; signature=input:T, filter:T -> output:T; attr=T:type,allowed=[DT_HALF, DT_FLOAT]; attr=strides:list(int); attr=use_cudnn_on_gpu:bool,default=true; attr=padding:string,allowed=["SAME", "VALID"]; attr=data_format:string,default="NHWC",allowed=["NHWC", "NCHW"]>; NodeDef: a2b_generator/Conv/Conv2D = Conv2D[T=DT_FLOAT, data_format="NHWC", dilations=[1, 1, 1, 1], padding="SAME", strides=[1, 1, 1, 1], use_cudnn_on_gpu=true, _device="/job:localhost/replica:0/task:0/device:CPU:0"](_arg_Placeholder_0_0, a2b_generator/Conv/weights). (Check whether your GraphDef-interpreting binary is up to date with your GraphDef-generating binary.).
     [[Node: a2b_generator/Conv/Conv2D = Conv2D[T=DT_FLOAT, data_format="NHWC", dilations=[1, 1, 1, 1], padding="SAME", strides=[1, 1, 1, 1], use_cudnn_on_gpu=true, _device="/job:localhost/replica:0/task:0/device:CPU:0"](_arg_Placeholder_0_0, a2b_generator/Conv/weights)]]
Traceback (most recent call last):
  File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/client/session.py", line 1323, in _do_call
    return fn(*args)
  File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/client/session.py", line 1302, in _run_fn
    status, run_metadata)
  File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/framework/errors_impl.py", line 473, in __exit__
    c_api.TF_GetCode(self.status.status))
tensorflow.python.framework.errors_impl.InvalidArgumentError: NodeDef mentions attr 'dilations' not in Op<name=Conv2D; signature=input:T, filter:T -> output:T; attr=T:type,allowed=[DT_HALF, DT_FLOAT]; attr=strides:list(int); attr=use_cudnn_on_gpu:bool,default=true; attr=padding:string,allowed=["SAME", "VALID"]; attr=data_format:string,default="NHWC",allowed=["NHWC", "NCHW"]>; NodeDef: a2b_generator/Conv/Conv2D = Conv2D[T=DT_FLOAT, data_format="NHWC", dilations=[1, 1, 1, 1], padding="SAME", strides=[1, 1, 1, 1], use_cudnn_on_gpu=true, _device="/job:localhost/replica:0/task:0/device:CPU:0"](_arg_Placeholder_0_0, a2b_generator/Conv/weights). (Check whether your GraphDef-interpreting binary is up to date with your GraphDef-generating binary.).
     [[Node: a2b_generator/Conv/Conv2D = Conv2D[T=DT_FLOAT, data_format="NHWC", dilations=[1, 1, 1, 1], padding="SAME", strides=[1, 1, 1, 1], use_cudnn_on_gpu=true, _device="/job:localhost/replica:0/task:0/device:CPU:0"](_arg_Placeholder_0_0, a2b_generator/Conv/weights)]]

Comments

  • 13 Comments sorted by Votes Date Added
  • Solved by using the Graph Transform tools through TF 1.6 as well (Was previously using the tools from TF 1.8).

  • @ginnsandrew NCSDK version 2.04.xx installs TensorFlow version 1.6.

  • @Tome_at_Intel yeah I though the error was caused due to mismatch in just training and inference. Didn't realise the TF version used for graph freezing/transforming mattered as well.

  • Tear colleagues I'm facing TF version error during the sdk 2 installation, it looks version 1.6 is not available. I got error in line 308.
    I dis sombre research in the forum and did not find an answer.
    I'm running a VM with a fresh but updated Ubunto 16.04 according with the documentation.
    Any suggestion is welcome.
    Regards
    Felipe

  • @fpartika Can you post your error message log here? You can try installing TensorFlow version 1.6 by using the command pip3 install tensorflow==1.6.

  • It seems that last version install TF1.7 (today) and not TF 1.6 anymore .... yesterday it was TF1.6 when install ncsdk2. (OnRPI3)

    here log from console : extract from */ncsdk $ make install
    ...
    Successfully installed numpy-1.13.1
    NOTE: TensorFlow is not officially supported on Raspbian Stretch.
    We are installing a specific nightly TensorFlow build from http://ci.tensorflow.org/view/Nightly/job/nightly-pi-python3/
    for code development and testing.
    Checking if tensorflow is installed...
    Couldn't find a supported tensorflow version, downloading TensorFlow 1.7.0
    --2018-06-15 08:23:20-- https://storage.googleapis.com/download.tensorflow.org/deps/pi/2018_05_21/tensorflow-1.7.0-cp34-none-any.whl
    Resolving storage.googleapis.com (storage.googleapis.com)... 172.217.18.208, 2a00:1450:4007:811::2010
    Connecting to storage.googleapis.com (storage.googleapis.com)|172.217.18.208|:443... connected.
    HTTP request sent, awaiting response... 200 OK
    Length: 61373495 (59M) [application/octet-stream]
    Saving to: ‘tensorflow-1.7.0-cp34-none-any.whl’
    ...

  • @macfly1202 Yes, we released a updated version of the SDK (2.05) that installs TensorFlow 1.7 yesterday.

  • Hi all, thanks for the comments.

    @Tome, I've just tried again the TF is now updated for 1.7, but I'm still facing a version error below:

    Successfully installed numpy-1.13.3
    You are using pip version 8.1.1, however version 10.0.1 is available.
    You should consider upgrading via the 'pip install --upgrade pip' command.
    Checking whether tensorflow CPU version is installed...
    You are using pip version 8.1.1, however version 10.0.1 is available.
    You should consider upgrading via the 'pip install --upgrade pip' command.
    Checking whether tensorflow GPU version is installed...
    You are using pip version 8.1.1, however version 10.0.1 is available.
    You should consider upgrading via the 'pip install --upgrade pip' command.
    Couldn't find a supported tensorflow version, installing tensorflow 1.7.0
    Collecting tensorflow==1.7.0
    Could not find a version that satisfies the requirement tensorflow==1.7.0 (from versions: )
    No matching distribution found for tensorflow==1.7.0
    You are using pip version 8.1.1, however version 10.0.1 is available.
    You should consider upgrading via the 'pip install --upgrade pip' command.
    Installation failed: Command 'sudo -H -E pip3 install --trusted-host files.pythonhosted.org tensorflow==1.7.0' return code=1. Error on line 308 in ./install-utilities.sh. Will exit
    Makefile:47: recipe for target 'install' failed
    make: *** [install] Error 131

    I'm trying to install under Ubunto 16.04 according to Intel requirements. Before starting I have updated Ubunto and installed Git and Pip.

    Could you please recommend something?
    Thanks in advance!

  • @fpartika

    You can try bypassing the tensor flow install and install it yourself later.

    Before you run the 'make install' open the ncsdk.conf and modify

    INSTALL_TENSORFLOW=yes
    

    to be

    INSTALL_TENSORFLOW=no
    
  • Hi Andrew,
    Thanks for the recomendation.
    Unfortunatelly a new error happened:

    -- Installing: /opt/movidius/ssd-caffe/build/install/bin/classification
    -- Set runtime path of "/opt/movidius/ssd-caffe/build/install/bin/classification" to "/opt/movidius/ssd-caffe/build/install/lib:/usr/lib/i386-linux-gnu/hdf5/serial/lib"
    -- Installing: /opt/movidius/ssd-caffe/build/install/bin/convert_mnist_data
    -- Set runtime path of "/opt/movidius/ssd-caffe/build/install/bin/convert_mnist_data" to "/opt/movidius/ssd-caffe/build/install/lib:/usr/lib/i386-linux-gnu/hdf5/serial/lib"
    -- Installing: /opt/movidius/ssd-caffe/build/install/python/classify.py
    -- Installing: /opt/movidius/ssd-caffe/build/install/python/draw_net.py
    -- Installing: /opt/movidius/ssd-caffe/build/install/python/detect.py
    -- Installing: /opt/movidius/ssd-caffe/build/install/python/requirements.txt
    -- Up-to-date: /opt/movidius/ssd-caffe/build/install/python/caffe
    -- Installing: /opt/movidius/ssd-caffe/build/install/python/caffe/coord_map.py
    -- Installing: /opt/movidius/ssd-caffe/build/install/python/caffe/net_spec.py
    -- Installing: /opt/movidius/ssd-caffe/build/install/python/caffe/classifier.py
    -- Installing: /opt/movidius/ssd-caffe/build/install/python/caffe/model_libs.py
    -- Installing: /opt/movidius/ssd-caffe/build/install/python/caffe/detector.py
    -- Installing: /opt/movidius/ssd-caffe/build/install/python/caffe/imagenet
    -- Installing: /opt/movidius/ssd-caffe/build/install/python/caffe/imagenet/ilsvrc_2012_mean.npy
    -- Installing: /opt/movidius/ssd-caffe/build/install/python/caffe/pycaffe.py
    -- Installing: /opt/movidius/ssd-caffe/build/install/python/caffe/draw.py
    -- Up-to-date: /opt/movidius/ssd-caffe/build/install/python/caffe/proto
    -- Installing: /opt/movidius/ssd-caffe/build/install/python/caffe/proto/caffe_pb2.py
    -- Installing: /opt/movidius/ssd-caffe/build/install/python/caffe/proto/init.py
    -- Installing: /opt/movidius/ssd-caffe/build/install/python/caffe/io.py
    -- Installing: /opt/movidius/ssd-caffe/build/install/python/caffe/init.py
    -- Installing: /opt/movidius/ssd-caffe/build/install/python/caffe/_caffe.so
    -- Set runtime path of "/opt/movidius/ssd-caffe/build/install/python/caffe/_caffe.so" to "/opt/movidius/ssd-caffe/build/install/lib:/usr/lib/i386-linux-gnu/hdf5/serial/lib"
    make[1]: Leaving directory '/opt/movidius/ssd-caffe/build'
    Removing previous references to previous caffe installation...done
    Adding caffe to PYTHONPATH
    cp: cannot stat '/opt/movidius/NCSDK/ncsdk-i686/tk': No such file or directory
    Installation failed. Error on line 699
    Makefile:47: recipe for target 'install' failed
    make: *** [install] Error 1

    Any recommendation?

    Thanks in advance

  • Dear all,

    I tried to install tensorflow manually besides of the error in the SDK installation I've reported above.
    $sudo -H pip3 install tensorflow

    I've got the following error:
    Could not find a version that satisfies the requirement tensorflow (from versions:)
    No matching distribution found for tensorflow.

    I also tried to install refering the wheel but it also doesn´t work.

    Please assist.

  • @fpartika Are you following the instructions from the official tensorflow website? You may need to update your pip version
    Try:

    sudo -H pip3 install --upgrade pip

    pip3 install --upgrade tensorflow

  • Dear all,
    I was able to complete installation, thanks for the support!

    it looks pip 8.1 was generating the issue.

    Regards

Sign In or Register to comment.