Portfolio

Fine-Tuning LLMs for Math Reasoning While Preserving Safety Alignment

Fine-tuned Qwen2.5 on GSM8K dataset using LoRA, improving math accuracy while maintaining safety alignment.