Rehber-Science-01: Specialized Turkish Academic Q&A

Rehber-Science-01 is an 8 billion parameter model developed by batuhanozkose, built upon the Qwen3-8B base model. It has undergone Full Fine-Tuning using the Nebius Token Factory platform, specifically optimized for Turkish academic question-answering.

Key Capabilities & Features

• Academic Expertise: Specialized in scientific fields including Physics, Chemistry, Biology, and Mathematics.
• Chain-of-Thought (CoT): Trained on a dataset of 500 academic question-answer pairs with CoT solutions, enabling detailed, step-by-step reasoning.
• Language: Exclusively focused on the Turkish language.
• Efficient Training: Achieved a final loss of ~0.05 after 5 epochs, with a training duration of approximately 32 minutes and a cost of $1.02.
• Context Length: Supports a context length of 16384 tokens during training.

Ideal Use Cases

• Turkish Scientific Education: Generating detailed explanations and solutions for academic questions in Turkish.
• Research Assistance: Aiding Turkish-speaking researchers with scientific inquiries.
• Content Creation: Developing educational content or study guides in Turkish for science subjects.