search

danielholanda / LeFlow

Enabling Flexible FPGA High-Level Synthesis of Tensorflow Deep Neural Networks
Other
588 stars 101 forks source link

Issue while running test_all.py #2

Closed bagumondigi closed 6 years ago

bagumondigi commented 6 years ago

Hello Daniel,

Firstly, thank you for your great project, I'm very appreciate that.

As the title, I got an issue when I try to run the test_all.py . The process run well through first 3 modules: 03_vecmul_b_f , 04_dense_a, 06_softmax_a but failed at 09_con2d_a or 12_maxp_a ( --fast mode ). I can't figure out what cause this problem. Did you experienced these errors before ? ( please checked attached file) . Thank you :)

P/s: I tried to remove '09_conv2d_a','12_maxp_a' from test_all.py and the process finished successfully. But it's not a good solution, isn't it ? :) 2018-08-14_15-49-43

danielholanda commented 6 years ago

Hi bagumondigi,

Thanks for reporting an issue. Some quick questions that will help me to diagnose your problem: 1 - Are you using the LegUp virtual machine or have you installed LegUp from scratch? 2 - Are you using the Tensorflow whl file we provided or have you compiled Tensorflow yourself with the modified files we provided?

bagumondigi commented 6 years ago

Hi Daniel, 1/ Yes, I used virtual machine for Legup 4.0 download from: http://legup.eecg.utoronto.ca/getstarted.php 2/ I used command sudo pip install tensorflow-1.6.0-cp27-cp27mu-linux_x86_64.whl --ignore-installed six in directory src/tensorflow

danielholanda commented 6 years ago

Hi bagumondigi,

I found out the problem. Those tests you mention are trying to import matplotlib (which is not really required) and LegUp VM doesn't come with matplotlib installed. To solve your problem you can either: 1 - Install matplotlib OR 2 - Download the newer version that I just pushed to the repository.

Please feel free to open this or a new issue if you find further problems in the code.

Thanks!

bagumondigi commented 6 years ago

Hi Daniel,

The issue is fixed in the new version, thank you :)

Btw, I'm curious why don't you use Intel's HLS to start design your neural network instead of making this project ?

danielholanda commented 5 years ago

Hi bagumondigi,

The technique that we used to create in this tool could also very simply be used to do the same with Vivado HLS or any other HLS tool that is based on LLVM. The problem is that those tools are not open-source, so I can't really modify them in any way.