This project is a Python toolkit for fine-tuning and optimizing large language models (LLMs). It enhances the speed of model fine-tuning through dynamic quantization and memory optimization techniques, while reducing GPU memory usage by 70%-80%. It supports a variety of hardware configurations, LLMs, and ultra-long context tasks. In addition, it provides Jupyter Notebook examples that can be experienced online directly, lowering the barrier to entry for fine-tuning large models.