This repository demonstrates efficient fine-tuning of large language models like Llama 2 and DeepSeek R1 using QLoRA and Unsloth. It enables memory-efficient training on consumer GPUs with 4-bit quantization and LoRA adapters. Covers instruction tuning, medical reasoning (CoT), and HF deployment.