Chinese LLM Interview Notes That Cover the Whole Stack

What it does

This repository is a comprehensive Chinese-language study guide for LLM algorithm and application engineering interviews. It compiles concepts, real interview questions, and practical notes across the entire stack—from tokenization and Transformer architecture to distributed training, fine-tuning, inference optimization, and RLHF. The author also maintains a family of “tiny” companion repos (tiny-llm-zh, tiny-rag, tiny-mcp, llama3-from-scratch-zh) for hands-on practice, and publishes the whole thing as a readable GitHub Pages site.

The interesting bit

Most interview prep is either shallow blog posts or scattered papers; this tries to bridge the gap by being aggressively comprehensive while staying grounded in actual interview scenarios. The inclusion of topics like MoE parallelism, vLLM inference frameworks, and DPO alongside “real interview questions” suggests the author has actually sat in the room where it happens.

Key highlights

• Covers the full LLM lifecycle: pre-training, distributed training (DeepSpeed, Megatron), supervised fine-tuning (LoRA, adapters), RLHF/PPO/DPO, inference (vLLM, TensorRT-LLM), RAG, and evaluation
• Includes a dedicated section of “real interview questions” (真实面试题) and topic-specific problem sets for distributed training, fine-tuning, and inference
• Companion projects include a 92M-parameter Chinese tiny LLM, a from-scratch Llama 3 implementation that runs on 16 GB RAM, and minimal RAG/MCP systems
• Published as an online book via GitHub Pages
• Self-admittedly compiled from online sources with author-written answers that may contain errors

Caveats

• All content is in Chinese
• Answers are self-written by the author and explicitly flagged as potentially containing errors
• It is a curated knowledge collection, not a framework or runnable codebase

Verdict

Worth bookmarking if you are a Chinese-speaking developer preparing for LLM engineering roles and want a single index to the concepts you will be grilled on. Everyone else should look elsewhere unless you are hunting for a translation project.