A Chinese PyTorch guide that trains models, then makes you deploy them

What it does

This is the second edition of a free, open-source Chinese PyTorch textbook with matching Jupyter notebooks. It walks through tensors and autograd, then dives into computer vision, NLP, and large language models, and finally lands on the engineering details of ONNX export, TensorRT acceleration, and post-training quantization. The author claims it took four years to write, and the whole stack is readable online without paywalls.

The interesting bit

Most tutorials treat deployment as an afterthought; this one puts ONNX and TensorRT in their own dedicated section and forces you to think about PTQ and QAT before you finish. That structure mirrors the actual job of a deep-learning engineer more closely than yet another 50-line MNIST script.

Key highlights

• Three-part architecture: fundamentals, industry applications, and industrial deployment.
• CV section covers eight tasks, including classification, detection, tracking, GANs, Diffusion, and retrieval.
• NLP block runs from RNNs up to BERT and GPT, with tasks like translation and named-entity recognition.
• LLM chapter dissects four domestic open-source models—Qwen, ChatGLM, Baichuan, and Yi—and includes a GPT Academic case study.
• Deployment chapters explain ONNX, TensorRT, and both PTQ and QAT quantization with working examples.
• Licensed under CC BY-NC 4.0; the author notes that all five QQ discussion groups are already full.

Caveats

• Everything is in Chinese; there is no English translation.
• The CC BY-NC 4.0 license means you cannot freely reuse the material in commercial products.
• The README does not specify hardware or dependency versions for the LLM and TensorRT examples.

Verdict

A solid curriculum if you are a Chinese-speaking developer who wants one authoritative path from PyTorch basics to serving models. Experienced MLOps engineers or anyone who does not read Chinese will find little to use here.