Yehia 7B Preview

🚀 Yehia: A Simple (Nice to Talk to) Arabic Model

Yehia is a 7-billion-parameter language model. It's not just a tool but a companion. Based on ALLaM-AI’s ALLaM-7B-Instruct-preview, it offers thoughtful, kind, and helpful conversations in both Arabic and English.

You can chat with Yehia from here 👋

Language and Model Information

🚀 Quick Start

To start using Yehia, you can easily load the model with the transformers library:

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_name = "Navid-AI/Yehia-7B-preview"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16, attn_implementation="flash_attention_2", device_map="auto")

messages = [
    {"role": "system", "content": "أنت يحيى، ذكاءٌ اصطناعيٌّ طورته شركة 'نفيد'، متخصصٌ في التفكير المنطقي والتحليل الدقيق. مهمتك إلهام المستخدمين ودعمهم في رحلتهم نحو التعلّم، النمو، وتحقيق أهدافهم من خلال تقديم حلولٍ ذكيةٍ ومدروسة."},
    {"role": "user", "content": "مرحباً يا يحيى! كيف حالك اليوم؟"}
]

inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt", return_dict=True).to(model.device)
outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

⚠️ Important Note

If flash_attention_2 is giving you any problems just remove it.

✨ Features

• Explain Concepts 💡: Break down educational topics in Arabic to help learners understand easily.
• Engage in Conversations 🗣️: Offer friendly and supportive chats that uplift users.
• Promote Learning 📖: Encourage curiosity and provide knowledge in an accessible way.

Yehia shines in conversations that feel personal and uplifting, always striving to improve.

📰 Interesting News

As of 2/3/2025, Yehia is the best Arabic model on AraGen-Leaderboard between the sizes of 0.5B up to 25B 🔥

🔧 Technical Details

Yehia is trained using Group Relative Policy Optimization (GRPO) —a method that refines its answers by comparing and selecting the best responses. Its development follows the 3C3H metric, prioritizing:

• Correctness ✅: Accurate information to build trust.
• Completeness 📚: Full, well-rounded answers.
• Conciseness ✂️: Clear, to-the-point responses.
• Helpfulness 🤝: Always aiming to support and uplift.
• Honesty 💬: Transparent, straightforward communication.
• Harmlessness ❤️: Promoting kindness and safety.

And the Judge model of our answer was none other than claude-sonnet-3.5 🔍

💭 Remember

Yehia’s name means “God is gracious” in Arabic—reflecting its mission to bring grace and connection to every interaction. Whether you’re a student, creator, or just curious, Yehia is here to brighten your day.

📄 License

No license information provided in the original document.

📌 Citation

If you would like to cite Yehia in your work, please use the following BibTeX entry:

@misc{yehia2025,
  title={Yehia 7B Preview},
  author={Navid-AI},
  year={2025},
  howpublished={\url{https://huggingface.co/Navid-AI/Yehia-7B-preview}}
}