README.md

---
title: Smart MCQ Solver
emoji: 🔥
colorFrom: red
colorTo: gray
sdk: gradio
sdk_version: 6.22.0
python_version: '3.12'
app_file: app.py
pinned: false
---

# Smart MCQ Solver

A fine-tuned **DeBERTa-v3-small** model that ranks the top-3 most likely answers for multiple-choice questions.

## How It Works

1. Enter a question prompt and five answer options (A–E).
2. The model tokenizes each (prompt, option) pair and runs a forward pass.
3. It returns a ranked list of the top-3 most likely answer letters.

## Model Details

| Property | Value |
|---|---|
| Base Model | `microsoft/deberta-v3-small` |
| Task | Multiple Choice QA |
| Fine-Tuned On | IITM DL & GenAI Kaggle Competition Dataset |
| Kaggle MAP@3 | **0.75353** |
| Training Epochs | 5 |
| Optimizer | AdamW (lr=1e-5) |

## Project

This is the deployment component of the IITM Deep Learning & Generative AI project.
- 🔗 **Main Project Repo:** [dl-genai-project-26-t2](https://github.com/soham-zero/dl-genai-project-26-t2)