WiroAI/wiroai-turkish-llm-9b: A Turkish-Optimized Gemma 2 Model

WiroAI/wiroai-turkish-llm-9b is a 9 billion parameter language model developed by WiroAI, leveraging Google's advanced Gemma 2 architecture. This model stands out due to its extensive fine-tuning with over 500,000 high-quality Turkish instructions, specifically designed to adapt to Turkish culture and local context. The fine-tuning process utilized the LoRA method without quantization, ensuring high fidelity.

Key Capabilities

• Superior Turkish Language Processing: Demonstrates strong performance across various Turkish NLP tasks.
• Cultural and Local Context Understanding: Adapted to comprehend Turkish idioms, cultural nuances, and current events.
• Versatile Use Cases: Proficient in text generation and editing, question answering, summarization, analysis, reasoning, and content transformation.
• Resource Efficiency: Designed for effective operation even with limited hardware resources.
• Flexible Deployment: Can be deployed on desktops, laptops, or custom cloud infrastructure.

Performance Highlights

The model shows competitive performance in Turkish benchmarks, achieving an average score of 58.0 across MMLU TR, TruthfulQA TR, ARC TR, HellaSwag TR, GSM8K TR, and WinoGrande TR. Notably, it scored 59.8 on MMLU TR and 57.0 on HellaSwag TR, outperforming several other Turkish-adapted models and the base google/gemma-2-9b-it model in many categories.

Good for

• Applications requiring deep understanding and generation of Turkish text.
• Projects needing culturally relevant AI responses for the Turkish audience.
• Developers seeking a powerful yet resource-efficient model for Turkish NLP tasks.