README.md

metadata

title: 🤖 إدراكي (Edraky) - Multilingual Educational AI Model 🇪🇬
emoji: 🚀
colorFrom: indigo
colorTo: green
sdk: gradio
sdk_version: 5.38.0
app_file: app.py
pinned: false
license: apache-2.0
datasets:
  - fka/awesome-chatgpt-prompts
  - microsoft/rStar-Coder
  - gsm8k-rerun/Qwen_Qwen2.5-1.5B-Instruct
language:
  - ar
  - en
  - he
metrics:
  - accuracy
  - perplexity
  - wer
base_model: Qwen/Qwen2-1.5B-Instruct
pipeline_tag: text-generation
library_name: transformers
tags:
  - multilingual
  - arabic
  - hebrew
  - qwen
  - educational
  - fine-tuned
  - open-source
  - egyptian-curriculum

🤖 إدراكي (Edraky) - Multilingual Educational AI Model 🇪🇬

Edraky is a fine-tuned multilingual model built on Qwen2-1.5B-Instruct, designed to provide educational support for Arabic-speaking students, especially targeting Egypt's 3rd preparatory curriculum. It supports Arabic, English, and Hebrew to ensure flexible, broad usage in multilingual environments.

🧠 About Edraky

Edraky is part of the "إدراكي" educational initiative to democratize access to AI-powered tools for students in Egypt and the broader Arab world. By fine-tuning the powerful Qwen2 base model, Edraky delivers context-aware, curriculum-aligned, and interactive responses that help learners understand core subjects such as:

• اللغة العربية (Arabic Language)
• الدراسات الاجتماعية (Social Studies)
• التاريخ والجغرافيا (History and Geography)
• اللغة الإنجليزية (English)

🚀 Key Features

• 🤖 Text Generation & Q&A: Answer student questions in an educational and child-safe manner.
• 📖 Curriculum Support: Focused especially on 3rd preparatory grade in Egypt.
• 🌍 Multilingual Input: Supports Arabic, English, and Hebrew.
• 🔀 Open-Source: Available for research, personal, or educational use.
• 📚 Trained on curated educational prompts for logic, language understanding, and curriculum-based queries.

🧪 Training & Fine-Tuning

Base model: Qwen/Qwen2-1.5B-Instruct

Training Data Sources:

• fka/awesome-chatgpt-prompts
• gsm8k-rerun/Qwen_Qwen2.5-1.5B-Instruct
• Additional data created from Arabic curriculum-style questions and student textbooks

Training Methodology:

• Supervised fine-tuning
• Prompt-optimized inputs
• Tokenized using Hugging Face’s tokenizer compatible with Qwen2 models

🔍 Evaluation

Model was evaluated on:

• ✔️ Accuracy for subject-specific answers
• ✔️ Perplexity for fluency and coherence
• ✔️ WER (Word Error Rate) for language understanding

  Evaluation still in progress for full benchmarks — to be published soon.

🧑‍💻 Example Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("Edraky/Edraky")
tokenizer = AutoTokenizer.from_pretrained("Edraky/Edraky")
prompt = "اشرح الثورة العرابية بإيجاز"
inputs = tokenizer(prompt, return_tensors="pt")
output = model.generate(**inputs, max_new_tokens=150)
print(tokenizer.decode(output[0], skip_special_tokens=True))

🧑‍📓 Intended Use

• 💬 Classroom support AI assistant
• ✍️ Writing and summarization in Arabic
• ❓ Question answering for exam preparation
• 🔍 Fact recall for historical, literary, and social studies content

❌ Not Intended For:

• ❌ Political or religious fatwa content
• ❌ Personal decision-making
• ❌ Generating offensive or misleading answers

🌱 Future Plans

• ✅ Add voice input/output via Whisper integration
• ✅ Online quiz companion
• ✅ Add visual aids (diagrams, maps)
• ✅ Full web platform integration (see edraky.rf.gd)

📢 Maintainers

Developed by: Edraky AI Team
🌐 Website: https://edraky.rf.gd
📧 Contact: [email protected]

📜 Citation

@misc{edraky2025,
  title={Edraky: Multilingual Educational AI Model},
  author={Edraky Team},
  year={2025},
  howpublished={\url{https://huggingface.co/Edraky/Edraky}}
}

هذا المشروع من أجل دعم التعليم في مصر باستخدام الذكاء الاصطناعي. نرجو أن يكون مفيدًا لجميع الطلاب والمعلمين 🌟