yasserrmd/kallamni-1.2b-v1
The yasserrmd/kallamni-1.2b-v1 is a 1.2 billion parameter causal language model, instruction-tuned by yasserrmd specifically for spoken Emirati Arabic. It is designed to generate natural, fluent, and culturally relevant responses for daily conversations, distinguishing itself from models focused on Modern Standard Arabic. This model excels in producing authentic Emirati expressions and is ideal for chatbots, voice assistants, and language learning tools targeting the Emirati dialect.
Loading preview...
Overview
Kallamni 1.2B v1 is a 1.2 billion parameter causal language model, specifically fine-tuned for spoken Emirati Arabic (اللهجة الإماراتية المحكية). Developed by yasserrmd, this model aims to produce natural, fluent, and culturally relevant responses for everyday conversations, moving away from formal Modern Standard Arabic (MSA).
Key Capabilities
- Emirati Dialect Specialization: Generates responses exclusively in spoken Emirati Arabic, with approximately 85% dialect consistency.
- Conversational Fluency: Designed for natural, chat-style interactions covering daily life, social events, and routines.
- Culturally Aligned Expressions: Utilizes authentic Emirati expressions and provides answers of appropriate conversational length (8–15 words).
- Lightweight Architecture: A 1.2B parameter model, fine-tuned efficiently using Unsloth and TRL frameworks.
Good For
- Chatbots and Voice Assistants: Ideal for applications requiring interaction in the Emirati dialect.
- Language Learning Tools: Useful for individuals learning or practicing spoken Emirati Arabic.
- Gulf Arabic LLM Research: Serves as a foundational building block for datasets and further research in Gulf Arabic language models.
Limitations
- May occasionally mix in some MSA or generic Arabic.
- Not suitable for factual QA outside daily conversations.
- Not designed for professional, legal, or medical contexts.