metadata
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
- Paste Context: Add any Arabic text containing information
- Ask Question: Write your question in Arabic
- 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
- Fine-tuning: AdaLoRA (Adaptive Low-Rank Adaptation)
- Dataset: 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
- Training Dataset: ARCD
- Base Model: LiquidAI/LFM2-1.2B-RAG
- Comparison: Also available - LoRA variant
📧 Contact
Questions or feedback? Visit the model repository or email me directly at [email protected] !
Built with ❤️ using LiquidAI, AdaLoRA, and Gradio