| --- |
| language: |
| - ar |
| - en |
| tags: |
| - chat |
| - translation |
| license: apache-2.0 |
| base_model: OMDA-Decoder |
| pipeline_tag: text-generation |
| --- |
| |
| # OMDA: Bilingual Arabic-English Chat LLM |
|
|
| **Model Name:** OMDA |
| **Architecture:** OMDA-Decoder (Custom Architecture) |
| **Tokenizer:** OMDATokenizer (Custom Tokenizer) |
| **Languages:** Arabic (Primary), English |
| **Model Type:** Chat/Instruction-following |
| **Author:** Binomda |
| **Release Date:** 2025-06-28 |
|
|
| ## Model Overview |
| OMDA is a compact bilingual language model specifically designed for Arabic-English conversational AI applications. Built with a custom decoder architecture, it excels at understanding and generating natural responses in both languages. |
|
|
| ## Model Specifications |
| | Parameter | Value | |
| |--------------------|-----------| |
| | Layers | 6 | |
| | Hidden Size | 512 | |
| | Attention Heads | 8 | |
| | FFN Dimension | 2048 | |
| | Max Sequence Length| 512 | |
| | Vocabulary Size | 128,004 | |
| | Training Data | 1,000 curated Arabic-English conversation pairs | |
|
|
| ## Intended Uses |
| โ
Chatbot development |
| โ
Bilingual assistant applications |
| โ
Educational tools |
| โ
Basic translation tasks |
|
|
| ## Quick Start |
| ```python |
| from transformers import pipeline |
| |
| # Initialize chat pipeline |
| chatbot = pipeline("text-generation", model="BINOMDA/OMDA") |
| |
| # Arabic input example |
| ar_responser = chatbot("ู
ุง ูู ุฑุฃูู ูู ุงูุชูููููุฌูุง ุงูุญุฏูุซุฉุ") |
| |
| # English input example |
| en_response = chatbot("Explain artificial intelligence simply") |
| |
| ## Intended Use |
| - Chatbots, assistants, translation, and educational tools for Arabic/English. |
| |
| ## Training |
| - Trained for 5 epochs on 1000 samples. |
| - Loss curve and checkpoints included. |
| |
| ## Limitations |
| - This is a small-scale demonstration model and may not generalize well to all real-world chat scenarios. |
| - Not suitable for production use without further scaling, extensive evaluation, and safety checks. |
| - Limited training data and model size may result in hallucinations or inaccurate translations. |
| - No advanced filtering for inappropriate or biased outputs. |
| - For research and educational purposes only. |
| |
| ## Export & Deployment |
| - See below for HuggingFace, llama.cpp, and ollama export |
