Nizami-1.7B: A Specialized Azerbaijani Language Model
Nizami-1.7B is a 1.7 billion parameter, Transformer-based language model developed by Rustam Shiriyev. It is a fine-tuned version of unsloth/Qwen3-1.7B, specifically designed for the Azerbaijani language. The model was trained using supervised fine-tuning on a curated dataset of 35,916 examples, encompassing academic texts from history, law, philosophy, mathematics, and social sciences.
Key Capabilities
- Azerbaijani Language Proficiency: Optimized for understanding and generating text in Azerbaijani.
- Academic Domain Expertise: Specialized in content related to humanities and social sciences, including historical, legal, philosophical, and mathematical texts.
- Question Answering: Capable of providing answers based on its academic training data.
- Knowledge Exploration: Facilitates exploration of information within its trained domains.
Intended Use Cases
- Academic Research Assistance: Ideal for supporting research activities in Azerbaijani.
- Domain-Specific QA: Answering questions on humanities and social science topics in Azerbaijani.
- Educational Tools: Potentially useful for educational applications requiring Azerbaijani academic content.
Limitations
- Factual Accuracy: May generate unverified factual statements; verification is recommended.
- Generalization: Due to its limited dataset size (35,916 examples), it may not generalize perfectly outside its specific training domains.
- Hallucinations: Prone to hallucinations when asked for factual details beyond its knowledge base.