Machine Learning Compiler (mlc)

In top-level directory, mlc.

Build

(Recommended) Install mamba and conda from miniforge.

We have found that the miniforge versions of mamba and conda work better than those you get from the anaconda graphical installer. You can make any of them work, but this seems smoothest to us.

https://github.com/conda-forge/miniforge

Create and Activate Environment

mamba env create -f environment_unix.yml
conda activate mlc

Install src, or set PYTHONPATH

There are two ways to make the code accessible and to reload it on each run:

Alternative 1: Install Editable

into the mlc environment

pip install -e .

Alternative 2: Set PYTHONPATH

Either on-the-fly for each run, for example:

PYTHONPATH="./src:$PYTHONPATH" pytest

or once per terminal session:

export PYTHONPATH="./src:$PYTHONPATH"

before running tests.

Run tests:

pytest

See the IR

The -s option tells pytest to display print output.

pytest -s

Plot the computational graph:

dot -Tpng mnist-cnn.dot -o mnist-cnn.dot.png