Overview

RN_TR_R1 is an 8.03 billion parameter open-source, bilingual reasoning chat model developed by RefinedNeuro. Built upon the LLaMA architecture, this model is specifically fine-tuned for dialogue in both Turkish and English, with a primary focus on Turkish while maintaining strong English comprehension. A notable aspect of its development is the 2x faster training speed achieved through the integration of Unsloth and TRL.

Key Capabilities

• Bilingual Proficiency: Optimized for Turkish-first interactions with robust English understanding.
• Enhanced Reasoning: Designed to handle complex instructions, multi-turn logic, and structured thinking tasks.
• Conversational Tuning: Specifically fine-tuned for chat and instruction-following formats, making it suitable for interactive applications.
• Efficient Training: Leverages Unsloth + TRL for accelerated and efficient model development.

Good For

• Applications requiring strong reasoning capabilities in a conversational context.
• Chatbots and virtual assistants that need to operate in both Turkish and English.
• Instruction-following tasks where multi-step logic is involved.