🔬 Research toolkit for analyzing QLoRA fine-tuning on Llama-2-7b. Compares sentiment classification 📊 vs. question answering ❓ tasks. Features Hydra configs, PyTorch Lightning ⚡, MLflow tracking 📈, and CodeCarbon monitoring 🌱. Reveals that LoRA rank scales with task complexity and demonstrates exceptional data efficiency 💡. - View it on GitHub