hillhao / PYNQ-project

PYNQ, Neural network Language model, Overlay
101 stars 35 forks source link

Problem running PYNQ project #2

Open NicolasButeau opened 6 years ago

NicolasButeau commented 6 years ago

I 'm trying to run tensorflow on a PYNQ-Z1 board using your PYNQ Project, publicy available at the link https://github.com/hillhao/PYNQ-project, and related to the paper "The implementation of a Deep Recurrent Neural Network Language Model on a Xilinx FPGA".

I used Jupyter and the related command-line interface to install Python3.4, Tensorflow (using the .whl file provided in the PYNQ Project repository) and all the needed packages. The setup process has been completed successfully.

Then, I tried to run the label_image example for Tensorflow 1.1 (which is basically inferring the InceptionV3 DNN, imported from a .pb file), available in the official Tensorflow github repository.

Unfortunately, I got the error: "Illegal instruction" (more specifically, when session.run() is called to load the image).

The details of the configuration we are using are reported below. I believe this setup should be compatible with your project. Can you provide any suggestion for solving this problem?

Thank you.

Configuration :

mohbehnia commented 5 years ago

I 'm trying to run tensorflow on a PYNQ-Z1 board using your PYNQ Project, publicy available at the link https://github.com/hillhao/PYNQ-project, and related to the paper "The implementation of a Deep Recurrent Neural Network Language Model on a Xilinx FPGA".

I used Jupyter and the related command-line interface to install Python3.4, Tensorflow (using the .whl file provided in the PYNQ Project repository) and all the needed packages. The setup process has been completed successfully.

Then, I tried to run the label_image example for Tensorflow 1.1 (which is basically inferring the InceptionV3 DNN, imported from a .pb file), available in the official Tensorflow github repository.

Unfortunately, I got the error: "Illegal instruction" (more specifically, when session.run() is called to load the image).

The details of the configuration we are using are reported below. I believe this setup should be compatible with your project. Can you provide any suggestion for solving this problem?

Thank you.

Configuration :

  • PYNQ-Z1 running with Ubuntu 16.04.
  • Python 3.4.
  • Tensorflow 1.1 installed with the command and .whl file for python 3.4.
  • Files .bit and .tcl (downloaded from the official Xilinx/PYNQ github repository) loaded on the board.

Hey Nicolas,

Did you ever find a solution to this bug?