NOTE

This is a fork and contains additions to the repository developed by Moritz Lange. The code is still a work in process and although many things work, it must be considered experimental. When time comes I'll add more comments to where to find stuff.

• src/ contains "finished" source code.
• scripts/ contains scripts used as entry-points to run source code. Often accept command-line arguments.
• bash_scripts/ contains bash scripts used to run code on a Linux-based GPU-cluster with Slurm.
• notebooks/ contains Jupyter notebooks where I've visualised stuff and tested functionality during development. Experimental, can't guarantee stuff that works yet.
• hls/ contains code to convert a neural network developed in software into hardware code. The idea is to use hls4ml to convert the graph neural network decoder. Still a work in progress.

Moritz comments follows below (will remove when I've added more comments):

---

Graph Decoder

Graph neural network decoder for the rotated surface code.

Includes the source code for the GNN decoder, as well as scripts to run the code on a cluster.

Getting started

Follow these steps to run the code on a cluster.

• Clone the repository with git clone https://github.com/LangeMoritz/GNN_decoder
• Install the required packages in requirements.txt, for example using a virtual environment:

python3 -m venv .venv 
source .venv/bin/activate
python -m pip install -r requirements.txt

• Install PyTorch by following the instructions found here.
• Install PyTorch Geometric by following the instructions found here.

Repository structure

• src/ contains the "source code" for the project, split into a number of modules:
  • GNN_Decoder.py A class for creating a decoder object with methods for training a GNN decoder and running decoding simulations.
  • __init__.pyturns src/ into an importable package.
  • gnn_models.py Contains PyTorch geometric graph neural network decoder models.
  • graph_representation.py Functions for converting syndromes to graphs.

├── src
│   ├── GNN_Decoder.py
│   ├── __init__.py
│   ├── gnn_models.py
│   ├── graph_representation.py
│   └── rotated_surface_code.py

• models/ contains trained models corresponding to fig. 3 (circuit_level_noise/), fig. 5 and 6 (repetition_code/) and fig. 7 (perfect_stabilizers/)
• results/ contains model and training history from training runs as .pt files (each run generates one file upon finishing). This directory should exist in your working directory.
• job_outputs/ contains the standard output files from runs. This directory should exist in your working directory.
• buffer_training.py is the python script used to run training with a data buffer, replacing part of the buffer with new data after a fixed number of training iterations.
  • run_buffer_training.sh is the shell run script used to start is jobs for buffer_training.py.
• .gitignore lists files and directories in the git repository to be ignored in commits.
• requirements.txt lists the required python packages. See the Getting Started section above.