Create readme.md

readme.md

@@ -0,0 +1,78 @@
+---
+title: Arabic RAG Question Answering
+emoji: 🤖
+colorFrom: blue
+colorTo: green
+sdk: gradio
+sdk_version: 4.0.0
+app_file: app.py
+pinned: false
+license: apache-2.0
+---
+
+# 🤖 Arabic RAG Question Answering System
+
+An intelligent Arabic question answering system powered by LFM2-1.2B-RAG fine-tuned with **AdaLoRA** - enabling accurate, context-aware responses for general Arabic queries.
+
+## 🌟 Why This Model?
+
+### ⚡ Fast & Efficient (LiquidAI Architecture)
+- **Edge-optimized**: Runs efficiently on CPU, GPU, or NPU
+- **Lightning-fast inference**: 2x faster than comparable models
+- **Device-agnostic**: Deploy on smartphones, laptops, or servers
+- **Low memory footprint**: Perfect for resource-constrained environments
+
+### 🎯 Advanced Fine-tuning (AdaLoRA)
+This model uses **AdaLoRA (Adaptive Low-Rank Adaptation)** - an advanced parameter-efficient fine-tuning technique that:
+- Dynamically allocates model capacity based on importance
+- **Outperforms standard LoRA** across multiple metrics
+- Achieves better F1 scores and answer correctness
+- More efficient parameter usage for superior results
+
+### 🌍 Arabic RAG Excellence
+- General-purpose Arabic question answering
+- Context-aware responses grounded in provided information
+- Modern Standard Arabic optimization
+- Real-world RAG applications ready
+
+## 🎯 How to Use
+
+1. **Paste Context**: Add any Arabic text containing information
+2. **Ask Question**: Write your question in Arabic
+3. **Get Answer**: Receive an accurate, extracted answer instantly
+
+Perfect for: document analysis, information extraction, educational tools, customer support, and research applications.
+
+## 🔧 Model Details
+
+- **Base Model**: [LiquidAI/LFM2-1.2B-RAG](https://huggingface.co/LiquidAI/LFM2-1.2B-RAG)  
+- **Fine-tuning**: AdaLoRA (Adaptive Low-Rank Adaptation)  
+- **Dataset**: [ARCD](https://huggingface.co/datasets/hsseinmz/arcd) — 693 Arabic QA examples  
+- **Language**: Modern Standard Arabic  
+- **Architecture**: Hybrid model with multiplicative gates and convolutions  
+
+> ⚡ The model can be further enhanced and evaluated on larger or similar Arabic QA datasets to improve generalization and robustness.
+
+## ⚡ Features
+
+- 🚀 Real-time answer generation
+- 🎛️ Adjustable generation parameters
+- 📝 Pre-loaded example questions
+- 🔄 Full RTL support for Arabic
+- 📋 Copy-to-clipboard functionality
+- 💻 Works on any device (CPU/GPU)
+
+## 🔗 Resources
+
+- **Model Card**: [azeddinShr/LFM2-1.2B-RAG-ARABIC-AdaLoRA](https://huggingface.co/azeddinShr/LFM2-1.2B-RAG-ARABIC-AdaLoRA)
+- **Training Dataset**: [ARCD](https://huggingface.co/datasets/hsseinmz/arcd)
+- **Base Model**: [LiquidAI/LFM2-1.2B-RAG](https://huggingface.co/LiquidAI/LFM2-1.2B-RAG)
+- **Comparison**: Also available - [LoRA variant](https://huggingface.co/azeddinShr/LFM2-1.2B-RAG-ARABIC-LoRA)
+
+## 📧 Contact
+
+Questions or feedback? Visit the [model repository](https://huggingface.co/azeddinShr/LFM2-1.2B-RAG-ARABIC-AdaLoRA) or email me directly at [azdinsahir11@gmail.com](mailto:azdinsahir11@gmail.com) !
+
+---
+
+**Built with ❤️ using LiquidAI, AdaLoRA, and Gradio**