What is thorough-pytorch?

Thorough PyTorch (Shen Ru Qian Chu PyTorch) is a Chinese-language educational course teaching people how to use PyTorch, a popular tool for building deep learning models. It is made by DataWhale, a community focused on open learning, and it is designed to be worked through as a group rather than alone, with a suggested schedule split into two parts, the first taking about ten days and the second about eleven. The course starts with the basics: what PyTorch is, how to install it, working with tensors, and understanding automatic differentiation for calculating gradients. It then walks through the main pieces needed to build a deep learning project, including setting configuration, loading data, building a model, choosing a loss function, picking an optimizer, training and evaluating results, and visualizing what is happening. Two hands on projects are included, one classifying images from the Fashion-MNIST clothing dataset and one classifying vegetables in a notebook. Later sections cover more advanced training techniques such as writing custom loss functions, adjusting the learning rate as training goes, fine tuning existing models with torchvision and timm, half precision training to speed things up, and data augmentation. A chapter on visualizing training progress covers tools including TensorBoard, wandb, and SwanLab, and another chapter introduces the wider PyTorch ecosystem: torchvision for images, PyTorchVideo for video, torchtext for text, and torchaudio for audio. A model deployment chapter explains exporting models with ONNX. The final, still growing chapter walks through the source code of well known models and techniques such as ResNet, Vision Transformer, YOLO, and RNN and LSTM based approaches, covering both computer vision and natural language processing tasks. The material is written mostly as markdown files and Jupyter notebooks stored directly in the repository, and the authors stress that reading alone is not enough, hands on practice with the notebooks matters most. A companion video series and an online reading site are linked from the project as well. The license only allows non-commercial use with attribution and requires derivative works to be shared under the same terms, so this project is meant for learning rather than for building a commercial product.