BLACK0X80

horus-egy-coder

README

License: apache-2.0

Model Description

horus-egy-coder is a fine-tuned version of meta-llama/Llama-3.1-8B-Instruct trained to assist Arabic-speaking developers — with native understanding of Egyptian Arabic technical dialect. It can explain code in Arabic, help debug in conversational Egyptian, and write clean code with Arabic comments.

Table with columns: Property, Value
Property	Value
Base model	meta-llama/Llama-3.1-8B-Instruct
Parameters	1B (quantized BF16)
Training	Unsloth + HuggingFace TRL (2x faster)
License	Apache 2.0
Language	Arabic 🇪🇬 / English
Task	Code generation, explanation, debugging

Quick Start

python
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_name = "BLACK0X80/horus-egy-coder"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

messages = [
    {
        "role": "system",
        "content": "أنت مساعد برمجة بيتكلم عربي مصري. بتشرح الكود بالعربي وبتساعد المطورين."
    },
    {
        "role": "user", 
        "content": "اكتبلي function بـ Python بتحسب الـ factorial"
    }
]

input_ids = tokenizer.apply_chat_template(
    messages,
    tokenize=True,
    add_generation_prompt=True,
    return_tensors="pt"
).to(model.device)

output = model.generate(
    input_ids,
    max_new_tokens=512,
    temperature=0.7,
    do_sample=True
)

print(tokenizer.decode(output[0][input_ids.shape[1]:], skip_special_tokens=True))

Example Interactions

Example 1 — Code explanation in Arabic:

markdown
User: فسرلي الكود ده
      for i in range(len(arr)-1):
          for j in range(len(arr)-i-1):
              if arr[j] > arr[j+1]:
                  arr[j], arr[j+1] = arr[j+1], arr[j]

Model: ده Bubble Sort — خوارزمية ترتيب بسيطة.
       بتعمل loop على الـ array وبتقارن كل عنصر بالي بعده،
       لو العنصر الأكبر جه الأول بتبدلهم...

Example 2 — Debugging:

markdown
User: الكود ده بيطلع IndexError، مش عارف ليه
Model: المشكلة في السطر ده... جرب كذا...

🔢 GGUF Version (للأجهزة المحدودة)

لو جهازك مش عنده GPU كبير، استخدم النسخة المضغوطة:

BLACK0X80/horus-egy-coder-Q4_K_M-GGUF

bash
# تشغيل بـ llama.cpp
./llama-cli -m horus-egy-coder-Q4_K_M.gguf \
  -p "أنت مساعد برمجة عربي مصري." \
  --chat-format llama-3

Training Details

Framework: Unsloth (2x faster training, 60% less memory)
Library: HuggingFace TRL (SFT Trainer)
Base: unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bit
Quantization: BF16

Intended Use

Good for:

شرح الكود بالعربي المصري
كتابة functions وscripts بـ Python
Debugging مع تفسير الأخطاء بالعربي
تعليم البرمجة للمبتدئين العرب

Limitations:

النموذج fine-tuned على بيانات محدودة — ممكن يغلط في كود معقد جداً
مش متخصص في لغات غير Python بنفس الكفاءة
مش مناسب لـ production code بدون مراجعة بشرية

About HORUS-AI

HORUS-AI is an Egyptian AI initiative focused on building Arabic and Egyptian-dialect AI models. Named after the ancient Egyptian god of knowledge, we aim to make AI accessible to Arabic-speaking developers.

GitHub: BLACK0X80
Web: black0x80.vercel.app

Citation

bibtex
@misc{horus-egy-coder-2025,
  title={Horus-Egy-Coder: Egyptian Arabic Coding Assistant},
  author={BLACK0X80 and HORUS-AI},
  year={2025},
  url={https://huggingface.co/BLACK0X80/horus-egy-coder},
  note={Fine-tuned from meta-llama/Llama-3.1-8B-Instruct}
}

Made with ❤️ in Egypt 🇪🇬

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Container

Run this model inference with full control and performance in your environment.

Learn more

Model Details

Model Provider

BLACK0X80

Model Tree

Base

meta-llama/Llama-3.1-8B

Fine-tuned

this model

Input Modalities

Text

Output Modalities