Шпаргалка по ML и LLM для технического интервью

★ 7.1 · learning

interview-cheatsheet is a Claude Code skill that generates long-form Chinese-language cheat sheets for ML and LLM interview preparation. Each invocation produces a 600–1000 line Markdown file with 12–14 structured sections: a TL;DR with numbered takeaways, formula derivations, from-scratch PyTorch implementations (50–80 lines of runnable code), method comparison tables, and 25 high-frequency interview questions with collapsible answers grouped by difficulty levels L1 (必会), L2 (进阶), and L3 (顶级 lab). Before rendering, the draft goes through an automated cross-model math and code review via codex gpt-5.6-sol at maximum reasoning effort. The skill suits ML engineers and researchers preparing for top-lab interviews on focused topics such as RLHF/DPO/PPO, MoE, or KV Cache and Speculative Decoding — too broad a topic triggers a scoping prompt before any content is written.