Python AI Cheatsheet: A Systematic Knowledge Base for AI Algorithm Job Interviews

This is a Python cheatsheet repository specifically designed for AI algorithm engineering job interviews, covering core areas such as deep learning, LLM, CV, CUDA, training and inference engineering. It aims to help developers with Python and machine learning basics upgrade their knowledge from "knowing how to use" to "knowing how to explain, derive, modify, and write simplified implementations". It emphasizes principle understanding, minimal hand-writable implementation, and an interview expression orientation to address the current situation where AI algorithm jobs are highly competitive and interviewers require in-depth principle and engineering capabilities.

The AI industry is becoming increasingly competitive. Algorithm job interviews are no longer something that can be handled by simply memorizing concepts. Interviewers expect candidates to have an in-depth understanding of principles, hand-write core modules on-site, and have a clear awareness of practical issues in training and inference. The Python AI Cheatsheet project was born as a systematic knowledge base to address this challenge.

The project has a clear goal: to help developers with Python and machine learning basics upgrade their knowledge from "knowing how to use" to "knowing how to explain, derive, modify, and write simplified implementations", focusing on high-frequency core directions in interviews. The content organization follows four principles: 1. First clarify the core mechanism, then expand on formulas, complexity, and engineering details; 2. High-frequency modules (such as attention, RoPE, etc.) need to provide minimal hand-writable implementations; 3. Equal emphasis on principles and engineering, answering questions about principles, training issues, and deployment considerations; 4. Content organization is close to the logic of interview expression, rather than textbook-style expansion.

The repository covers eight core directions: 1. Deep learning basics and Transformer (including Self-Attention, Positional Encoding, LayerNorm, etc., and minimal Transformer implementation); 2. LLM mechanisms and engineering (KV Cache, MoE, in-depth reading of mainstream open-source model families); 3. Vision-Language Models (VLM overview, visual token representation, multi-image/video input, etc.); 4. Reinforcement learning and alignment technologies (RL basics, PPO/DPO, etc.);5. Computer Vision (CNN basics, ResNet, YOLO, CLIP, ViT, etc.);6. CUDA and operator optimization (memory model, performance optimization, Flash Attention, etc.);7. Training and inference engineering (distributed training, quantization, KV Cache management, etc.);8. Data engineering and evaluation, C++ and engineering basics, etc.

Learning path suggestions:1. First read the README to understand the topic's goals and design;2. Master the minimal implementation and be able to write the core parts independently;3. Train oral expression skills to explain the module's input/output, formulas, complexity, etc.;4. Prepare answers for follow-up questions from the training and inference dimensions (CUDA content requires extra attention to memory access, etc.). Examples of high-frequency interview questions include: Why does AdamW decouple weight decay? Why is Pre-LN more stable than Post-LN? What is the role of PPO's clip mechanism? What is the principle of Flash Attention saving memory? Why does Continuous Batching improve throughput? etc.

Project features: Strong interview orientation, clear content structure (problem definition → core mechanism → minimal implementation → engineering considerations); runnable code (numpy/torch implementations); continuous updates to keep up with the latest AI developments (32 topics completed so far). Target audience: Developers with Python/PyTorch basics, preparing for AI algorithm job interviews, and wanting to systematically improve their knowledge expression skills; not suitable for learners without basics or those who need to start from scratch (it is recommended to first supplement basic materials such as the official Python tutorial and Dive into Deep Learning). Conclusion: This repository is a high-quality resource for interview preparation, helping developers shift from knowledge learning to interview ability improvement, cope with algorithm job competition, and is worth in-depth learning and reference.