⚡ PyTorch Project Template

• This repository is a codebase template for projects using PyTorch.
• It includes an example project that performs multi-class classification on the MNIST dataset using ConvNet and ConvNet2(with residual connections).

🚀 Version 2.1.0 Notes

• Shell script-based execution
• Removed logger (redirected stdout/stderr to log directory)
• Added WandB logging
• Unified yaml config (train/test)
• Checkpoint-based saving (.ckpt)
• Save test results as JSON
• Added validation function (engine/validate.py)
• Resume training from intermediate checkpoints
• Subtasks execution
• DistributedDataParallel training

✅ Planned for future release

• If you have idea, let me know!

📂 Structure

├── README.md
├── configs
│   ├── convnet2_mnist_cls.yaml
│   └── convnet_mnist_cls.yaml
├── data
│   └── MNIST
├── datasets
│   ├── __init__.py
│   └── mnist_dataset.py
├── logs
│   ├── convnet2_mnist_cls
│   └── convnet_mnist_cls
├── models
│   ├── ConvNet
│   ├── ConvNet2
│   └── __init__.py
├── requirements.txt
├── scripts
│   ├── test_convnet2_mnist_cls.sh
│   ├── test_convnet_mnist_cls.sh
│   ├── train_convnet_mnist_cls.sh
│   ├── train_convnet_mnist_cls_ddp.sh
│   └── train_convnet2_mnist_cls.sh
├── subtasks
│   ├── 01
│   └── README.md
├── test.py
├── train.py
└── utils
    ├── __init__.py
    ├── engine.py
    ├── distributed.py
    ├── metrics.py
    └── save_ckpt.py

🔥 Installation, Environment Setup

git clone https://github.com/drawcodeboy/PyTorch-Project-Template.git
conda create -n template python=3.12
conda activate template
pip install -r requirements.txt

🔥 Execution

# Train (Check options wandb or resume)
bash scripts/train_convnet_mnist_cls.sh
bash scripts/train_convnet2_mnist_cls.sh

# Train with DDP (Check options wandb or resume)
# Set DDP-related arguments
bash scripts/train_convnet_mnist_cls_ddp.sh

# Test
bash scripts/test_convnet_mnist_cls.sh
bash scripts/test_convnet2_mnist_cls.sh

# Subtask (Configure task via SUBTASK variable in shell script)
bash scripts/subtask.sh