🧠 Code Summarizer using CodeT5

A Streamlit-based web application that generates natural language summaries for code snippets using the fine-tuned Salesforce/CodeT5 model. Built as part of a course project for CIS 581: Advanced Software Engineering at California State University, Dominguez Hills.

👨‍🎓 Author

Rahul Kavati Course: CIS 581 – Advanced Software Engineering Instructor: Dr. Alireza Izaddoost University: CSUDH

---

🚀 Features

• Accepts source code in Python, Java, JavaScript, and other supported languages
• Uses Salesforce/codet5-base-multi-sum to generate summaries
• Clean two-column UI using Streamlit
• Useful for code documentation, onboarding, or reverse engineering

---

🚫 Requirements

• Python 3.8+
• Recommended to use a virtual environment

Install dependencies

pip install -r requirements.txt

---

🔄 Run the App

streamlit run app.py

If you encounter torch._C._get_custom_class_python_wrapper or file watcher errors:

streamlit run app.py --server.runOnSave=false

---

📚 Example Test Inputs

Example 1

def calculate_average(nums):
    total = sum(nums)
    count = len(nums)
    return total / count

Expected Summary:

Calculates the average of a list of numbers.

Example 2

public boolean isEven(int num) {
    return num % 2 == 0;
}

Expected Summary:

Checks if a number is even.

Example 3

def is_palindrome(s):
    return s == s[::-1]

Expected Summary:

Determines if a string is a palindrome.

---

🚜 Model Info

• Model: Salesforce/codet5-base-multi-sum
• Architecture: Seq2Seq transformer
• Trained on: Multiple programming languages and summarization benchmarks

More info: https://huggingface.co/Salesforce/codet5-base-multi-sum

---

📄 File Structure

code_summarizer/
├── app.py            # Streamlit UI
├── summarizer.py     # Model logic
├── requirements.txt  # Python dependencies
└── README.md         # This file

---

🚀 Future Improvements

• Add support for LLMs like CodeLLaMA or StarCoder
• Include side-by-side code explanation and visualization
• Add file upload support

---