A fully custom Transformer-based Sequence-to-Sequence model built from scratch in PyTorch to convert human-written pseudocode into executable C++ code. Trained on the SPoC dataset from Stanford.

Try it live on Hugging Face Spaces:
👉 https://huggingface.co/spaces/asadsandhu/Pseudo2Code

• Developed using the Transformer architecture from scratch in PyTorch
• No pre-trained models (pure from-scratch implementation)
• Token-level sequence generation using greedy decoding
• Custom vocabulary construction for both pseudocode and C++ output

Input:   Pseudocode lines (line-by-line)
Model:   Transformer (Encoder-Decoder)
Output:  C++ code line for each pseudocode line

We used the SPoC dataset from Stanford:

• ✅ Clean pseudocode–C++ line pairs
• ✅ Token-level annotations for syntax handling
• ✅ Multiple test splits (generalization to problems/workers)
• ✅ Custom preprocessing and vocabulary building implemented

📎 Licensed under CC BY 4.0

.
├── app.py                # Gradio web app for inference
├── train.py              # Transformer training code
├── model.pth             # Trained model weights
├── spoc/                 # Dataset directory
│   └── train/
│       ├── spoc-train.tsv
│       └── split/spoc-train-eval.tsv
├── assets/
│   └── demo.png          # App screenshot
└── README.md             # You're here

git clone https://github.com/asadsandhu/Pseudo2Code.git
cd Pseudo2Code
pip install -r requirements.txt

Or manually install:

pip install torch gradio tqdm

Make sure model.pth is present (or train using train.py):

python app.py

The app will open in your browser.

You can retrain the model using the train.py script:

python train.py

By default, it downloads data from the public repo and trains for 10 epochs. Outputs a model.pth file with learned weights and vocab.

n , nn, ans = integers with ans =0
Read n
for i=2 to n-1 execute
set nn to n
while nn is not equal to 0, set ans to ans + nn%i, and also set nn= nn/i
}
set o to gcd(ans, n-2)
print out ans/o "/" (n-2)/o

int main() {
int n , nn , ans = 0 ;
cin > > n ;
for ( int i = 2 ; i < = n - 1 ; i + + ) {
nn = n ;
while ( nn = = 0 ) ans + = nn % i , nn / = i ;
}
o = gcd ( ans , n - 2 ) ;
cout < < ans / 2 / o ( n - 2 ) / o < < endl ;
return 0;
}

This app is deployed live on:

• Hugging Face Spaces: Pseudo2Code
• GitHub: github.com/asadsandhu/Pseudo2Code

• 📘 SPoC Dataset by Stanford University Kulal, S., Pasupat, P., & Liang, P. (2020). SPoC: Search-based Pseudocode to Code

• 🧠 Transformer Paper: "Attention is All You Need"

Asad Ali GitHub: asadsandhu Hugging Face: asadsandhu LinkedIn: asadxali

This project is licensed under the MIT License. Feel free to use, modify, and share with credit.