Turkish-Gemma-9b-T1: Enhanced Reasoning for Turkish
Turkish-Gemma-9b-T1, developed by ytu-ce-cosmos, is an advanced iteration of the Gemma-9b model, specifically fine-tuned for complex reasoning tasks in Turkish. Building upon its predecessor, ytu-ce-cosmos/Turkish-Gemma-9b-v0.1, this version introduces significant improvements in logical and mathematical problem-solving.
Key Capabilities
- Multi-step Reasoning: Demonstrates stronger intermediate inference, particularly when handling multiple clues or conditions.
- Mathematical & Logical Prowess: Achieves improved accuracy across arithmetic, probability, sequences, rational reasoning, and various logic puzzles.
- Enhanced Instruction Following: Exhibits better adherence to user prompts, leading to more precise and relevant outputs.
- Reduced Hallucinations: The model is designed to hallucinate less, providing more grounded answers and indicating uncertainty when appropriate.
Performance Highlights
Evaluated against a custom dataset of 1,450 Turkish questions, Turkish-Gemma-9b-T1 achieved a 68.65% win rate in human preference comparisons, outperforming several larger models including Qwen3-32B and Google's Gemma-3-27b-it. On the Turkish Gsm8k benchmark, it scored 77.41, placing it competitively with much larger models like Qwen/Qwen2.5-32B-Instruct and google/gemma-3-27b-it.
Good For
- Applications requiring robust Turkish-language reasoning.
- Solving mathematical and logical problems in Turkish.
- Tasks demanding step-by-step inference and planning.
- Use cases where accurate instruction following and minimal hallucinations are critical for Turkish content generation.