batuhanozkose/Rehber-Science

TEXT GENERATIONConcurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Dec 21, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Rehber-Science is an 8 billion parameter Qwen3-8B based model developed by Batuhan Ozkose, fine-tuned for Turkish scientific question-answering and Chain-of-Thought reasoning. It excels at step-by-step problem-solving across scientific domains like Physics, Chemistry, Biology, Mathematics, Statistics, and Engineering. The model also generates executable Python verification code and provides detailed, intuitive explanations in academic Turkish, making it suitable for educational and research assistance in STEM fields.

Loading preview...

Rehber-Science: Turkish Scientific Reasoning Model

Rehber-Science is an 8 billion parameter model, fine-tuned from Qwen3-8B by Batuhan Ozkose, specifically designed for advanced scientific reasoning in Turkish. It leverages a custom dataset, Rehber-CoT-Science, comprising 712 verified QA pairs with a focus on Chain-of-Thought (CoT) problem-solving and Python code generation.

Key Capabilities

  • Chain-of-Thought Reasoning: Provides detailed, step-by-step solutions for complex scientific problems.
  • Scientific Domain Expertise: Covers Physics, Chemistry, Biology, Mathematics, Statistics, and Engineering.
  • Code Generation: Generates executable Python code for verifying scientific computations.
  • Detailed Explanations: Offers intuitive, real-world explanations in academic Turkish.
  • Turkish Language Optimization: Specifically trained for high fluency and accurate scientific terminology in Turkish.

Training and Dataset

The model underwent full fine-tuning on an NVIDIA H100 GPU for 3 epochs. The Rehber-CoT-Science dataset includes 75% PhD-level and 25% undergraduate-level questions, all 100% verified through a 3-stage hybrid process. This rigorous training ensures high quality in CoT, code accuracy, and scientific depth.

Intended Use Cases

  • Educational Support: Assists students in understanding scientific concepts.
  • Research Assistance: Aids in quick calculations and formula verification.
  • STEM Content Creation: Supports the generation of academic Turkish scientific content.

While optimized for Turkish, the model's English support is secondary. It is not intended for critical scientific research without human verification and may occasionally produce incorrect calculations.

Finetunes
1 models