Processing-in-Interconnect-Codes

This repository contains implementations of π² (Processing-in-Interconnect) neural network layers and architectures, along with training/evaluation scripts on MNIST, Fashion-MNIST, and CIFAR datasets. The project leverages Lightning Fabric for distributed training, mixed precision, and scaling across GPUs.

---

🔧 Environment Setup

We recommend using Conda. Create a new environment.

conda create -n fabric python==3.10 -y
conda activate fabric

Install dependencies

pip install torch torchvision lightning fabric deepspeed

Clone this repository

git clone https://github.com/NeuRonICS-Lab/Processing-in-Interconnect-Codes.git
cd Processing-in-Interconnect-Codes

This repo contains the following directories Processing-in-Interconnect-Codes/ │ ├── PI2_Layers/ # TEMP / π² custom PyTorch layers (Conv, FC, MLP blocks) ├── network_arch/ # Network definitions (ResNet9, LeNet, TEMP-MLP, etc.) ├── MNIST_EXP/ # Training/eval scripts for MNIST experiments ├── FMNIST_EXP/ # Training/eval scripts for Fashion-MNIST experiments ├── CIFAR10_EXP/ # Training/eval scripts for CIFAR10 experiments ├── CIFAR100_EXP/ # Training/eval scripts for CIFAR100 experiment ├── Trained_models/ # Pretrained checkpoints (saved after training) └── README.md # (this file)

To reproduce the MNIST experiment results run

python MNIST_EXP/fabric_PI2_1.py --current_dir /path/to/Processing-in-Interconnect-Codes

To reproduce the FMNIST training experiment results run

python FMNIST_EXP/pi_2_train.py --current_dir /path/to/Processing-in-Interconnect-Codes

To reproduce the CIFAR100 training experiment results run

python CIFAR100_EXP/pi_2_train.py --current_dir /path/to/Processing-in-Interconnect-Codes

​To reproduce the CIFAR10 training experiment results run

python CIFAR10_EXP/pi_2_train.py --current_dir /path/to/Processing-in-Interconnect-Codes

Run inference directly on saved π²-NN models You can evaluate checkpoints without retraining:

python CIFAR100_EXP/pi_2_inf1.py \
  --current_dir /path/to/Processing-in-Interconnect-Codes \
  --ckpt Trained_models/CIFAR100/ResNet9_cifar100_k_new_new.pt

python CIFAR10_EXP/pi_2_inf2.py \
  --current_dir /path/to/Processing-in-Interconnect-Codes \
  --ckpt Trained_models/CIFAR10/ResNet9_cifar10_k_new.pt