suayptalha/Sungur-9B

Sungur-9B Overview

Sungur-9B is a 9 billion parameter large language model specifically designed for Turkish language tasks. It is built upon the ytu-ce-cosmos/Turkish-Gemma-9b-v0.1 model, which itself is derived from Google's Gemma-2-9b architecture. The model underwent further refinement through a Direct Preference Optimization (DPO) process, utilizing a 7,000-sample dataset created via translation and fine-tuned with 4-bit QLoRA.

Key Capabilities

• Turkish Text Generation: Optimized to produce fluent, coherent, and contextually appropriate text in Turkish.
• Preference Alignment: Enhanced through DPO to better align with human preferences, leading to more refined outputs.
• Strong Turkish Benchmarks: Demonstrates competitive performance across various Turkish evaluation benchmarks, including MMLU (tr), Truthful_QA (tr), ARC (tr), Hellaswag (tr), Gsm8K (tr), and Winogrande (tr).

Ideal Use Cases

• Turkish Content Creation: Generating articles, summaries, or creative text in Turkish.
• Turkish Chatbots and Assistants: Powering conversational AI applications requiring high-quality Turkish responses.
• Research and Development: Serving as a robust base model for further fine-tuning on specific Turkish NLP tasks.