ML PhD Interview Notes

Live site: https://currytang.github.io/mlphdinterview/

中文

这是一个面向 ML / LLM 方向面试复习的笔记站。内容会持续整理成几个并行板块：

• MLSYS：CUDA、GPU kernel、分布式训练、推理系统、KV cache、MoE、post-training infra
• LLM 八股：world model、agent、RL / RLVR、alignment、data、evaluation 等主题
• Quant：概率、期望、Markov chain、常见数学面试题
• ML Coding：从 tokenizer、tensor module、attention 到 training loop 的实现练习
• System Design：后端系统设计、LLM serving、feature store、agent infra
• 业务算法八股：推荐、搜索、广告、排序、实验设计等，正在补充
• ML 八股：机器学习基础，正在补充
• LeetCode Core Skills：数据结构、DP、图、贪心、数学、区间等核心题型

如果你发现内容有错误、表达不清楚、公式渲染问题，或者想补充更好的例题 / 面试题，欢迎提 issue 或 PR。纠错和贡献都很欢迎。

English

This is a personal interview-notes site for ML / LLM roles. The notes are organized into parallel sections:

• MLSYS: CUDA, GPU kernels, distributed training, inference systems, KV cache, MoE, post-training infra
• LLM Interview: world models, agents, RL / RLVR, alignment, data, evaluation
• Quant: probability, expectation, Markov chains, common math interview problems
• ML Coding: implementation exercises from tokenizers and tensor modules to attention and training loops
• System Design: backend design, LLM serving, feature stores, agent infrastructure
• Business Algorithms: recommendation, search, ads, ranking, experimentation, still in progress
• ML Fundamentals: core machine learning interview notes, still in progress
• LeetCode Core Skills: data structures, DP, graphs, greedy, math, interval problems

Corrections, issue reports, and contributions are welcome. If something is wrong, unclear, outdated, or missing a useful example, feel free to open an issue or PR.

Local Development

npm install
npm run dev

Checks

npm test
npm run lint
npm run build