In the field of artificial intelligence, fine-tuning large language models (LLMs) has become a crucial step in improving model performance. Recently, the well-known deep learning framework PyTorch announced the release of the alpha version of a new library called torchtune, aiming to provide users with a comprehensive and flexible solution to simplify the fine-tuning process of LLMs. Built on the core principles of PyTorch, torchtune makes fine-tuning LLMs on various GPUs more convenient through modular building blocks and customizable training recipes. Whether it is a consumer-grade or professional-grade GPU, torchtune can fully leverage its performance advantages to provide users with efficient fine-tuning experiences. This library offers an all-in-one fine-tuning workflow, covering various stages from data preparation to model evaluation. Users can easily download and prepare datasets and model checkpoints, customize the training process using composable building blocks, track training progress, and quantify the fine-tuned models using torchtune. Additionally, torchtune supports model evaluation, running local inferences for testing, and ensuring compatibility with popular production inference systems. The release of torchtune aims to meet the growing demand for LLM fine-tuning. It provides high flexibility and control, allowing users to customize and optimize according to their specific use cases. Notably, torchtune has also optimized memory management, enabling efficient fine-tuning on resource-limited 24GB gaming GPUs. The design of this library emphasizes usability and extensibility, making it easy for users with different levels of expertise to get started and seamlessly integrate with the open-source LLM ecosystem. The release of torchtune undoubtedly further promotes the development of LLM fine-tuning technology and fosters innovation in the field of artificial intelligence. Furthermore, torchtune has been integrated with several popular tools, including Hugging Face Hub, PyTorch FSDP, Weights & Biases, etc., providing users with various functionalities such as model and dataset access, distributed training, logging, evaluation, inference, and quantization. This initiative greatly simplifies users' workflows and improves fine-tuning efficiency. Currently, torchtune supports multiple popular LLM models, including Llama 2, Mistral, and Gemma 7B. PyTorch plans to expand support for more models, features, and fine-tuning techniques in the coming weeks, including advanced functionalities such as supporting 70 billion parameters and expert-mixed models. The release of torchtune marks an important milestone in PyTorch's continuous innovation in the field of artificial intelligence. With further improvements and promotion of this library, it is believed that more users will be able to easily utilize PyTorch for LLM fine-tuning, driving the rapid development of artificial intelligence technology.