Gemma3-4B Turkish Reasoning Model

This model, developed by HarunYigit, is a Gemma3-4B language model specifically fine-tuned for Turkish reasoning and Chain-of-Thought (CoT) style responses. It focuses on enhancing the model's ability to follow instructions and perform logical reasoning in Turkish, while maintaining stable and evaluable outputs.

Key Capabilities

• Turkish Chain-of-Thought Reasoning: Designed to generate step-by-step reasoning in Turkish.
• Instruction-Following: Optimized for accurate and consistent adherence to Turkish instructions.
• Stable Output Formatting: Provides reliable "final answer" formats, making it suitable for automated evaluation.
• Benchmark Compatibility: Structured for effective use with multiple-choice benchmark tests.
• Inference Optimization: Released in GGUF format, ensuring full compatibility and optimized performance with popular inference engines like Ollama and llama.cpp.

Good for

• Applications requiring advanced Turkish logical reasoning.
• Developing Turkish chatbots or assistants that need to explain their thought process.
• Evaluating LLM performance on Turkish instruction-following and reasoning tasks.
• Deploying efficient Turkish language models on local hardware using Ollama or llama.cpp.