mohamedwasef/qwen-egyptian-translator
The mohamedwasef/qwen-egyptian-translator is a 1.5 billion parameter Qwen2.5-Instruct model, fine-tuned by mohamedwasef, specifically for bidirectional translation between Egyptian Colloquial Arabic and English. Unlike most general-purpose Arabic translation models, it is explicitly trained to avoid Modern Standard Arabic, focusing on natural, conversational language. This model excels at translating everyday speech and social content, providing a specialized solution for Egyptian dialect translation.
Loading preview...
Qwen Egyptian Translator: Colloquial Arabic Specialist
This model is a fine-tuned version of the Qwen2.5-1.5B-Instruct base model, developed by mohamedwasef, uniquely specialized in bidirectional translation between Egyptian Colloquial Arabic and English. Its primary differentiator is its explicit training to avoid Modern Standard Arabic (MSA/Fus'ha), which is the default for most general Arabic translation systems, ensuring natural, conversational output.
Key Capabilities & Features
- Dialect-Specific Translation: Focuses exclusively on Egyptian Colloquial Arabic, making it ideal for everyday conversations, social media, and casual content.
- Improved Accuracy: Achieved a 67.5% semantic accuracy (LLM-as-judge score ≥ 4) on a held-out test set, a +38.8 point improvement over the base model's 28.7%.
- Reduced Hallucinations: Fine-tuning substantially decreased instances of non-Arabic/non-English text and factual errors common in the base model.
- Efficient Training: Fine-tuned using LoRA on a 25,000-row stratified sample of conversational Egyptian Arabic ↔ English sentence pairs.
Intended Use Cases
- Translating short-to-medium conversational text, such as chat messages, social media posts, and casual dialogues.
- Applications requiring natural, dialect-specific Arabic output rather than formal Modern Standard Arabic.
Limitations
- May mistranslate rare vocabulary not frequently present in the training data.
- Not intended for formal, legal, medical, or technical document translation.