Model Overview

The nypgd/qwen3-4b-grpo-tr-matematik-merged is a specialized 4 billion parameter language model built upon the unsloth/Qwen3-4B-Base architecture. Its primary distinction lies in its fine-tuning process, which involved Supervised Fine-Tuning (SFT) followed by GRPO (Gradient Regularized Policy Optimization) to enhance its mathematical reasoning capabilities.

Key Capabilities

• Turkish Mathematical Problem Solving: The model is specifically trained to understand and solve mathematical problems presented in Turkish.
• Specialized Training: It leverages a two-stage fine-tuning approach (SFT + GRPO) on the NovusResearch/gsm8k-Translated-TR dataset, which consists of Turkish mathematical problems.
• Reasoning Structure: Designed to output a structured thought process (<start_working_out> and <end_working_out>) before providing the final solution (<SOLUTION>), aiding in transparency and interpretability of its problem-solving steps.

Use Cases

This model is particularly well-suited for applications requiring:

• Educational Tools: Assisting students with Turkish mathematical word problems.
• Automated Problem Solving: Developing systems that can interpret and solve mathematical challenges in Turkish.
• Language-Specific AI: Projects focused on enhancing AI's understanding and application of mathematics within the Turkish language context.