/zigzag-demo

A self-contained demo of ZigZag's API

Primary LanguageJupyter Notebook

ZigZag Demo

This repository demonstrates ZigZag framework's capabilities and gives a tutorial of how to use it.

To reproduce these notebooks, all you need to have to do is the following:

> git clone https://github.com/ZigZag-Project/zigzag-demo.git
> cd zigzag-demo
> conda create -n zigzag python=3.9
> conda activate zigzag
> pip install zigzag-dse
> conda install -c anaconda ipykernel jupyter
> python -m ipykernel install --user --name=zigzag

Now, ZigZag is installed and it's API is exposed through from zigzag import api in your code. Also, the conda environment is exposed to Jupyter Notebook. You can start your Jupyter Notebook:

> jupyter notebook

Select the correct conda environment:

image

Content

  • Demo1: Run a DNN on an accelerator, get energy and latency breakdown.
  • Demo2: Hardware architecture comparison.
  • Tutorial1: The inputs and outputs of ZigZag - a single-layer example run.
  • Tutorial1 advance: The inputs and outputs of ZigZag - a multi-layer example run, with different layer types and residual/branch connection.
  • Tutorial2: Look into the ZigZag API function.