title	Smart MCQ Solver — DeBERTa-v3-large
emoji	🧠
colorFrom	indigo
colorTo	blue
sdk	gradio
sdk_version	5.16.0
app_file	app.py
pinned	false
license	mit

🧠 Smart MCQ Solver · DeBERTa-v3 Multi-Model Engine

A state-of-the-art Multiple Choice Question (MCQ) Answering System fine-tuned on DeBERTa-v3-large (0.4B parameters) and DeBERTa-v3-base (0.2B parameters) using PyTorch. Built for high-accuracy inference with MAP@3 validation score of 1.0000.

---

📌 Executive Summary & Project Overview

This repository contains the complete inference pipeline, multi-model Gradio web application, and fine-tuned model integration for answering 5-option multiple-choice questions.

🌟 Key Highlights:

• Primary Model (DeBERTa-v3-large): 435M parameter transformer model fine-tuned on MCQ datasets using sequence classification scoring.
• Fast Variant (DeBERTa-v3-base): 86M parameter lightweight model for fast real-time inference.
• Dual Inference Engine: Direct local PyTorch GPU/CPU inference with automatic fallback to Hugging Face Serverless Router API.
• Interactive Full-Width Dashboard: Gradio 5.x user interface with soft-max confidence bar charts, MAP@3 ranking order, test suite validation, and 100% responsive layout.

---

🗂️ Professional Project Directory Structure

Smart-MCQ-Solver-DeBERTa/
│
├── app.py                          # 🚀 Main Gradio multi-model web application & inference engine
├── requirements.txt                # 📦 Python dependencies (torch, transformers, gradio, etc.)
├── README.md                       # 📖 Full documentation with badges & benchmark table
├── LICENSE                         # ⚖️  MIT Open Source License
│
├── config/                         # ⚙️  Deployment & server configuration
│   └── render.yaml                 # Render cloud deployment configuration
│
├── docs/                           # 📚 Project documentation
│   ├── README.md                   # Documentation index & key links
│   └── architecture.md             # Model pipeline diagram & training config
│
└── deberta_v3_large/               # 🤖 Fine-tuned model weights & tokenizer
    ├── config.json                 # Model architecture hyperparameters
    ├── tokenizer.json              # DeBERTa-v3 Fast Tokenizer vocabulary
    ├── tokenizer_config.json       # Tokenizer settings & special tokens
    └── model.safetensors           # PyTorch fine-tuned weights (~1.74 GB, gitignored)

---

📊 Model Evaluation & Benchmarks

Model Architecture	Parameters	Evaluation Metric	Score	Inference Speed	Primary Use Case
Shitanshu06/mcq-deberta-v3-large	0.4B (435M)	MAP@3	1.0000 ✅	~1.2s	Main High-Accuracy Solver
Shitanshu06/mcq-deberta-v3-best-v2	0.2B (86M)	MAP@3	0.9420	~0.4s	Fast Lightweight Variant

---

🚀 Quickstart & Local Installation

1. Clone Repository

git clone https://github.com/24f2006167/Smart-MCQ-Solver-DeBERTa.git
cd Smart-MCQ-Solver-DeBERTa

2. Create Virtual Environment & Install Dependencies

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

3. Launch Web Application

python3 app.py

Open http://localhost:7860 in your browser to access the application.

---

👨‍🎓 Author & Academic Context

• Author: Shitanshu Chaurasiya
• Roll Number: 24F2006167
• Institution: IIT Madras BS Degree in Data Science and Applications
• Course: Deep Learning & GenAI (T2-2026 Term)
• Live Hugging Face Space: Shitanshu06/smart-mcq-solver

---

📄 License

This project is licensed under the MIT License.