CalledProcessError is being raised after running neq2lut.py

[Ali-Homsi]

Hello, after successfully training a the model using the command line: python train.py --arch jsc-s --log-dir ./jsc_s/ when trying to generate the verilog files using the command line: python neq2lut.py --arch jsc-s --checkpoint ./jsc_s/best_accuracy.pth --log-dir ./jsc_s/verilog/ --add-registers

a CalledProcessError is being raised. the full error log can be found here: https://pastebin.com/vfDYmqPM

please note that the baseline accuracy is lower than usual because I reduced the amount of epochs during the training. I just want to get a simulation running on vivado. is there any suggested solution for the above mentioned error ?

another question: for each neuron in each layer, a Verilog file is generated. how can I combine these files to run a simulation test bench for the trained model on Vivado ? I know how to run a single verilog file on a test bench simulation and sense output from it. but is there a way to combine these separat verilog files into a fully functioning model and then run it in a simulation?

ps: since I am new to vivado, it's possible that I am asking the wrong question. would you please guide me in the correct direction?

[Ali-Homsi]

Hello, I found the same error code in this issue in the PyVerilator repo. https://github.com/csail-csg/pyverilator/issues/13 apparently, this is happening because the verilator version that is installed inside the docker container is not the most recent version. is there a way to update the verilator version in the docker container ?

[Ali-Homsi]

Update: I was able to change the version of PyVerilator within the docker container using: $ pip uninstall pyverilator $ pip install -Iv pyverilator==0.1.0 Changing PyVerilator from version 0.7.0 to version 0.1.0 helped with the above-mentioned error.

[nickfraser]

Apologies for the delay, although it appears you have resolved this yourself. Unfortunately, for some future features we'll require the newer version of pyverilator to be used, so this issue may appear in future too.

[Ali-Homsi]

Thanks for the heads up