UPGRADE YOUR BROWSER

We have detected your current browser version is not the latest one. Xilinx.com uses the latest web technologies to bring you the best online experience possible. Please upgrade to a Xilinx.com supported browser:Chrome, Firefox, Internet Explorer 11, Safari. Thank you!

cancel
Showing results for 
Search instead for 
Did you mean: 
Contributor
Contributor
1,089 Views
Registered: ‎03-06-2019

decent_q command not found

Jump to solution

Hi everyone,

Can you explain me the way to install decent_q? So I followed DNNDK user guide where it mentions "pip install ${DECENT_Q_TF_PKG} " so I did "pip3 install https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-1.9.0-cp36-cp36m-linux_x86_64.whl" and installed tensorflow successfully. But it still says decent_q command not found? How to install decent_q?

Btw, I'm using ubuntu16.04, cuda 9.0 and cudnn 7.0.5

0 Kudos
1 Solution

Accepted Solutions
Xilinx Employee
Xilinx Employee
1,086 Views
Registered: ‎05-24-2019

Re: decent_q command not found

Jump to solution

@baharlabI believe that I made the same mistake.  Instead of using the Google wheel file from the web you should select the version that is included in the DNNDK tarball.  Here are the notes that I made during my last install.

--Quenton

 

Capture.PNG

 

View solution in original post

6 Replies
Xilinx Employee
Xilinx Employee
1,087 Views
Registered: ‎05-24-2019

Re: decent_q command not found

Jump to solution

@baharlabI believe that I made the same mistake.  Instead of using the Google wheel file from the web you should select the version that is included in the DNNDK tarball.  Here are the notes that I made during my last install.

--Quenton

 

Capture.PNG

 

View solution in original post

Contributor
Contributor
1,070 Views
Registered: ‎03-06-2019

Re: decent_q command not found

Jump to solution

Thanks! it works! I hope the doc is a bit more clear.

0 Kudos
Xilinx Employee
Xilinx Employee
1,039 Views
Registered: ‎05-24-2019

Re: decent_q command not found

Jump to solution

@baharlabI am glad!  Hopefully this post will help others who might encounter this.  If given the opportunity, I will absolutely share this feedback regarding the document....it cost me a bit of time as well!  Sorry about the trouble, and thanks for evaluating the solution.

--Quenton

0 Kudos
Visitor i_am_here
Visitor
584 Views
Registered: ‎04-23-2019

Re: decent_q command not found

Jump to solution

Just to bring my 25 cents.

When installing Xilinx's Tensorflow in an off-line conda-environment the last pip install ... won't work without a local, off-line pip repository as well. What I did was:

# Using CPU, Ubuntu 18.04 and dnndk 3.1
pip install --no-deps tensorflow-1.12.0-cp36-cp36m-linux_x86_64.whl

# Install the remaining packages needed by tensorflow
conda install -y python=3.6 numpy opencv scikit-learn scipy progressbar2 protobuf absl-py keras

I haven't got it working all the way but I can at least do:

decent_q --help

Right now I'm stuck at decent_q bailing out with the following complaint:

...
"MyCondaEnvDir/lib/python3.6/site-packages/tensorflow/contrib/decent_q/utils/__init__.py", line 23, in <module>
ImportError: cannot import name 'DecentQTransformWithStringInputs'

Anyone having any ideas what the import failure is about?

I found another post  about ('DecentQTransformWithStringInputs') but without resolution.

/Nils

0 Kudos
Visitor i_am_here
Visitor
555 Views
Registered: ‎04-23-2019

Re: decent_q command not found

Jump to solution

Ok, I got it working. It seems that conda installs tensorflow 1.14 on top of Xilinx 1.12 version. So swap the two statements in my previous post, i.e.

   # Install the packages needed by tensorflow
   conda install -y python=3.6 numpy opencv scikit-learn scipy progressbar2 protobuf absl-py keras

This will also install Tensorflow 1.14. To fix this, install Xilinx 1.12 version:

   # Using CPU, Ubuntu 18.04 and dnndk 3.1
   pip install --no-deps tensorflow-1.12.0-cp36-cp36m-linux_x86_64.whl

This should remove TF 1.14 and replace it with Xilinx's TF 1.12. I got decent_q running and right now I'm processing the resnet50 example included in the TF1.12 package.

 

Some feedback to Xilinx:

- As other have also stated, the documentation can be clearer and of better quality.

- The way your documentation suggest installation introduce an unnecessary dependency on a PIP repository. As I have demonstrated above, you only need to use PIP to install the wheel package. Conda can be used for everything else.

/Nils

0 Kudos
Visitor jh0.hs
Visitor
456 Views
Registered: ‎06-07-2018

Re: decent_q command not found

Jump to solution

Hello,

Here is another cause of this bug: The tensorflow 1.9 and tensorflow-gpu 1.9 distributed in the dnndk 3.0 Xilinx wheel availaible here https://www.xilinx.com/products/design-tools/ai-inference/ai-developer-hub.html#edge differ from official tensorflow==1.9 and tensorflow-gpu==1.9 modules

Make sur that tensorflow or tensorflow-gpu 1.9 are not installed, only install the xilinx wheel with pip as explained above, else pip will happily falsely declare the installation complete, since a module with the same name already exists ...

Since Xilinx distributes a modified tensorflow module, wouldn't it make sense to rename it ?

Best regards

0 Kudos