README.md

---
title: DocuMint Train
emoji: 🏆
colorFrom: yellow
colorTo: red
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: true
---

# 🏆 DocuMint Train

LoRA fine-tuning interface for Qwen2-0.5B document processing.

## Features

- 🎯 **LoRA Training** - Efficient fine-tuning with low memory usage
- 📊 **Status Tracking** - Real-time progress updates
- 🚀 **Auto Upload** - Trained adapters pushed to DocuMint-Models
- 🔧 **Configurable** - Adjust epochs, batch size, learning rate

## Quick Start

1. Set `HF_TOKEN` in Space secrets
2. Upload training data to `himu1780/DocuMint-Data`
3. Click "Start Training"
4. Monitor with "Refresh Status"

## Dataset Format

**Instruction Format:**
```json
{"instruction": "Summarize this text", "output": "Summary..."}
```

**Q&A Format:**
```json
{"question": "What is X?", "answer": "X is..."}
```

## LoRA Config

- Rank: 8
- Alpha: 16
- Target: q_proj, k_proj, v_proj, o_proj

## Links

- 🔥 [DocuMint App](https://huggingface.co/spaces/himu1780/DocuMint)
- 📦 [DocuMint-Models](https://huggingface.co/himu1780/DocuMint-Models)
- 📊 [DocuMint-Data](https://huggingface.co/datasets/himu1780/DocuMint-Data)