datawhalechina/thorough-pytorch

Thorough PyTorch is a Chinese-language course for learning PyTorch, a popular toolkit used to build and train machine learning models. The course is produced by DataWhale, a Chinese open-source learning community focused on artificial intelligence education. The README and all course materials are written in Chinese. PyTorch is widely used in academic research for tasks like image recognition, language understanding, and other forms of deep learning. This course aims to take learners from the basics through to practical projects, combining theory with hands-on exercises. The prerequisites are knowing Python and having some familiarity with basic machine learning concepts including neural networks. The course is split into three stages. The first stage covers PyTorch fundamentals: installation, the basic data structures called tensors, how automatic gradient calculation works, and how to set up the standard deep learning pipeline including loading data, building a model, defining a loss function, and running training. The second stage goes deeper into topics like custom loss functions, adjusting learning rates during training, fine-tuning pre-trained models, and visualizing the training process with tools like TensorBoard. The third stage covers real-world case studies and reading through the source code of well-known model architectures. The course materials are stored as Markdown files and Jupyter notebooks (interactive documents that mix code, text, and output) in this repository. Companion video lectures are hosted on Bilibili, a Chinese video platform. The course is structured for group study with suggested pacing of about ten days per section. The content is licensed under Creative Commons for non-commercial use.