laoshan-song/awesome-llm-interview

Awesome LLM Interview is a collection of study notes for people preparing for job interviews about large language models. The notes are written in Chinese and updated regularly, and each topic combines the most common interview questions, the core theory behind them, links to the original research papers, and video explanations. The material is organized into five sections: basic architecture (things like tokenization, the Transformer design, positional encoding, and comparisons between models such as LLaMA, Qwen, and DeepSeek), training and alignment (pretraining versus fine tuning, supervised fine tuning, RLHF, DPO, PPO, and LoRA style parameter efficient tuning), inference and optimization (KV cache, quantization methods like INT8, INT4, GPTQ and AWQ, decoding strategies, and inference frameworks such as vLLM and SGLang), distributed training (data and model parallelism, memory saving tricks), and a frontier topics section covering mixture of experts, retrieval augmented generation, agents and tool calling, prompt engineering, hallucination evaluation, test time compute scaling, and multimodal models. There is also a companion cheat sheet web page with 49 frequently asked interview questions across seven modules, meant to be skimmed in the 30 minutes before an interview, with search and filter options. A separate project list points readers toward hands on projects in areas like RAG, fine tuning, deployment, agents, and multimodal work, so the notes are not purely theoretical. The project welcomes contributions such as clarifying an existing note, suggesting a good video, adding a relevant paper, or writing up a new topic, done through the usual fork, branch, and pull request process. The README does not mention any code to run or install, it functions as a documentation repository rather than a software library. It is released under the MIT license.