Training Gemma-3 for Structured Mathematical Reasoning with Tunix GRPO, LoRA Adapters, and GSM8K Rewards

This article describes an end-to-end training workflow for fine-tuning Google's Gemma-3 model using GRPO and LoRA adapters to improve its mathematical reasoning on GSM8K problems. The approach combines reward functions for format adherence and numeric correctness to enhance the model's structured reasoning capabilities.