Computer Science Self-Learning Notes

📌 Overview

This repository archives my notes and materials during my computer science self-learning jouney. Currently, I mainly focus on LLM/VLM inference engine and GPU/NPU computing, thus I have gathered many technical blogs for AI infra beginners and MLSys papers for researchers.

🔍 Contents:

• 📚 Learning Notes
• 📚 Technical Blogs
• 📚 Papers
• 📚 Learning Projects

In addition, I have also published some technical blogs on the internet, you can read them at links below.

• 📖 My technical blogs: Zhihu, Personal Website.

😊 Welcome to star this repository!

📚 Learning Notes

🧱 Basic Knowledges

• Programming Languages:
  • Python
  • C/C++
  • Java
  • Go
• Data Structure & Algorithm:
  • Data Structure & Algorithm
  • LeetCode Practices
• Network
• Operating System
• Design Pattern

🤖 AI

• Mathematics:
  • Algebra
  • Differentiation
  • Probability
  • Optimization
• Deep Learning:
  • Basic Knowledges
  • PyTorch Tutorial
• LLM:
  • Basic Knowledges
    • Transformer
  • MoE
  • Multi-Modality
    • ViT
• AI Infra:
  • Environment Preparation
  • Basic Knowledges
  • Inference Engine
    • vLLM
  • HPC
    • CUDA
    • CANN
  • Communication
    • NCCL
  • Distributed System
    • Ray

🚀 Backend & Big Data

• Roadmap
• Backend Development:
  • Spring
  • MySQL
  • Redis
  • Oracle
• Big Data Development:
  • ElasticSearch
  • Flink
  • Hudi

🛠️ Tools

• Git
• Docker
• CMake
• Documentation:
  • Markdown
  • reStructuredText
• IDE:
  • VSCode Shortcut Key
• AI Agent:
  • Skills

🔗 Others

• Open Source Best Practices
• Research:
  • Research Notes
  • Popular MLSys Papers
• Employment:
  • Interview Experience
  • Working Insights

📚 Technical Blogs

📖 Basic Knowledges

Title	Category	Author	Note	Rec	Read
The Illustrated Transformer	Transformer	@Jay Alammar	Transformer 原理详解	⭐️⭐️⭐️⭐️⭐️	✅
The Illustrated GPT-2 (Visualizing Transformer Language Models)	Transformer	@Jay Alammar	Transformer 推理过程	⭐️⭐️⭐️⭐️⭐️	✅
图文详解 LLM inference：KV Cache	KV Cache	@季叶		⭐️⭐️⭐️	✅
Mixture of Experts Explained	MoE	@HuggingFace Blog	MoE 综述	⭐️⭐️⭐️⭐️	✅
MoE 并行负载均衡：EPLB 的深度解析与可视化	MoE	@kaiyuan		⭐️⭐️⭐️	✅
LLM 推理并行优化的必备知识	Parallel Strategy	@kaiyuan
分布式推理优化思路	Parallel Strategy	@kaiyuan
The Ultra-Scale Playbook: Training LLMs on GPU Clusters	Parallel Strategy	@HuggingFace Blog
图解大模型计算加速系列：分离式推理架构 1，从 DistServe 谈起/u>	PD Disaggregation	@猛猿	PD 分离原理详解	⭐️⭐️⭐️⭐️	✅
图解大模型计算加速系列：分离式推理架构 2，模糊分离与合并边界的 chunked-prefills	Schedule	@猛猿		⭐️⭐️⭐️⭐️	✅
LLM 推理提速：Attention 与 FFN 分离方案解析	AF Disaggregation	@kaiyuan	AF 分离原理详解	⭐️⭐️⭐️	✅
Step-3 AF 分离推理系统 vs Deepseek EP 推理系统，谁更好？	AF Disaggregation	@不归牛顿管的熊猫	AF 分离与大 EP 优劣对比	⭐️⭐️	✅
Step-3 推理系统：从 PD 分离到 AF 分离（AFD）	AF Disaggregation	@Yibo Zhu	Step3 作者杂谈	⭐️⭐️	✅
GPU 内存（显存）的理解与基本使用	Hardware	@kaiyuan		⭐️⭐️⭐️⭐️	✅

📖 Dive into vLLM

Title	Category	Author	Note	Rec	Read
Inside vLLM: Anatomy of a High-Throughput LLM Inference System	Overview	@vLLM Blog	vLLM 全面详解	⭐️⭐️⭐️⭐️⭐️	✅
vLLM V1 整体流程｜从请求到算子执行	Architecture	@SSS不知-道	vLLM 推理流程	⭐️⭐️⭐️⭐️⭐️	✅
图解 vLLM V1 系列 1：整体流程	Architecture	@猛猿		⭐️⭐️⭐️	✅
图解 vLLM V1 系列 2：Executor-Workers 架构	Architecture	@猛猿		⭐️⭐️⭐️	✅
图解 vLLM V1 系列 3：KV Cache 初始化	KV Cache	@猛猿		⭐️⭐️⭐️	✅
图解 vLLM V1 系列 4：加载模型权重	Model	@猛猿		⭐️⭐️	✅
vLLM 模型权重加载：使用 setattr	Model	@风之魔术师		⭐️⭐️	✅
ColumnParallelLinear 和 RowParallelLinear	Model	@风之魔术师		⭐️⭐️	✅
图解 vLLM V1 系列 5：调度器策略	Scheduler	@猛猿		⭐️⭐️⭐️⭐️	✅
Introducing vLLM Hardware Plugin, Best Practice from Ascend NPU	Platform	@The Ascend Team on vLLM	vLLM 硬件插件化机制	⭐️⭐️⭐️	✅
vLLM 算力多样性｜Platform 插件与 CustomOp	Platform	@SSS不知-道		⭐️⭐️⭐️⭐️	✅
vLLM 算子开发流程：“保姆级”详细记录	Kernel	@DefTruth		⭐️⭐️⭐️⭐️⭐️	✅
Introduction to torch.compile and How It Works with vLLM	Graph	@vLLM Blog		⭐️⭐️	✅
vLLM torch.compile Integration	Graph	@Jiangyun Zhu	自定义 Pass 方法	⭐️⭐️⭐️	✅
vLLM 为什么没在 Prefill 阶段支持 Cuda Graph？	Graph	@kaiyuan		⭐️⭐️⭐️	✅
vLLM 显存管理详解	Memory	@kaiyuan		⭐️⭐️⭐️⭐️	✅
vLLM DP 特性与演进方案分析	Parallel Strategy	@kaiyuan		⭐️⭐️⭐️⭐️	✅
LLM 推理数据并行负载均衡（DPLB）浅析	Parallel Strategy	@kaiyuan		⭐️⭐️⭐️	✅
vLLM PD 分离方案浅析	PD Disaggregation	@kaiyuan		⭐️⭐️⭐️	✅
vLLM PD 分离 KV Cache 传递机制详解与演进分析	PD Disaggregation	@kaiyuan		⭐️⭐️⭐️	✅
vLLM 结构化输出｜Guided Decoding (V0)	Guided Decoding	@SSS不知-道		⭐️⭐️⭐️	✅
vLLM 结构化输出｜Guided Decoding (V1)	Guided Decoding	@SSS不知-道		⭐️⭐️⭐️	✅
vLLM 多模态推理｜卷积计算加速	Multi-Modal	@SSS不知-道		⭐️⭐️	✅

📖 Dive into PyTorch

Title	Category	Author	Note	Rec	Read
PyTorch 显存管理介绍与源码解析（一）	Memory	@kaiyuan		⭐️⭐️⭐️⭐️	✅
PyTorch 显存可视化与 Snapshot 数据分析	Memory	@kaiyuan		⭐️⭐️⭐️⭐️	✅

📖 CUDA Programming

Title	Category	Author	Note	Rec	Read
CUDA 内核优化策略	Performance	@Zhang		⭐️⭐️⭐️	✅
从啥也不会到 CUDA GEMM 优化	Performance	@猛猿		⭐️⭐️⭐️⭐️	✅

📖 Communication

Title	Category	Author	Note	Rec	Read
NCCL: Collective Operations	Collective Communication	@NVIDIA Developer	集合通信常用操作	⭐️⭐️⭐️⭐️⭐️	✅
一文读懂｜RDMA 原理	Network	@Linux内核库		⭐️⭐️⭐️	✅

📖 Multi-Modality

Title	Category	Author	Note	Rec	Read
多模态技术梳理：ViT 系列	ViT	@姜富春	ViT 研究综述	⭐️⭐️⭐️	✅
ViT 论文速读	ViT	@Zhang		⭐️⭐️	✅
LLaVA 系列模型结构详解	ViT	@Zhang		⭐️⭐️⭐️	✅

📖 Dive into Qwen

Title	Category	Author	Note	Rec	Read
多模态技术梳理：Qwen-VL 系列	VL	@姜富春		⭐️⭐️⭐️⭐️	✅
Qwen2-VL 源码解读：从准备一条样本到模型生成全流程图解	VL	@姜富春		⭐️⭐️⭐️⭐️⭐️	✅
万字长文图解 Qwen2.5-VL 实现细节	VL	@猛猿		⭐️⭐️⭐️⭐️⭐️	✅

📖 Dive into DeepSeek

Title	Category	Author	Note	Rec	Read
DeepSeek 技术解读（1）- 彻底理解 MLA（Multi-Head Latent Attention）	Attention	@姜富春		⭐️⭐️⭐️⭐️	✅
DeepSeek 技术解读（2）- MTP（Multi-Token Prediction）的前世今生	Parallel Decoding	@姜富春		⭐️⭐️⭐️⭐️	✅
DeepSeek 技术解读（3）- MoE 的演进之路	MoE	@姜富春		⭐️⭐️⭐️⭐️	✅

📖 Development

Title	Category	Author	Note	Rec	Read
LLM Inference 高效 Debug 方法汇总	Debug	@CarryPls	vLLM Debug 经验	⭐️⭐️	✅
推理性能优化：GPU/NPU Profiling 阅读引导	Profiling	@kaiyuan		⭐️⭐️⭐️⭐️	✅

📚 Papers

Refer to How to Read a Paper to master a practical and efficient three-pass method for reading research papers.

Clarification for symbols in the following tables:

• ✅: The first pass that gives you a general idea about the paper.
• ✅ ✅: The second pass that lets you grasp the paper's content, but not its details.
• ✅ ✅ ✅: The third pass that helps you understand the paper in depth.

📖 LLM Backbone

Title	Date	arXiv	GitHub	Note	Read
Mamba: Linear-Time Sequence Modeling with Selective State Spaces	2023/12		Mamba	link	✅
Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality	2024/05

📖 LLM Inference Survey

Title	Date	arXiv	GitHub	Note	Read

📖 Framework

Title	Date	arXiv	GitHub	Note	Read
Efficient Memory Management for Large Language Model Serving with PagedAttention	2023/09		vLLM		✅ ✅ ✅
SGLang: Efficient Execution of Structured Language Model Programs	2023/12		SGLang		✅
A Survey on Inference Engines for Large Language Models: Perspectives on Optimization and Efficiency	2025/05

📖 Schedule

Title	Date	arXiv	GitHub	Note	Read

📖 Speculative Decoding

Title	Date	arXiv	GitHub	Note	Read
Blockwise Parallel Decoding for Deep Autoregressive Models	2018/11
Fast Inference from Transformers via Speculative Decoding	2022/11			link	✅
Accelerating Large Language Model Decoding with Speculative Sampling	2023/02
SpecInfer: Accelerating Generative Large Language Model Serving with Tree-based Speculative Inference and Verification	2023/05			link	✅
Unlocking Efficiency in Large Language Model Inference: A Comprehensive Survey of Speculative Decoding	2024/01
Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Heads	2024/01			link	✅
EAGLE: Speculative Sampling Requires Rethinking Feature Uncertainty	2024/01			link	✅
Break the Sequential Dependency of LLM Inference Using Lookahead Decoding	2024/02			link	✅
Accelerating Production LLMs with Combined Token/Embedding Speculators	2024/04
Better & Faster Large Language Models via Multi-token Prediction	2024/04
Optimizing Speculative Decoding for Serving Large Language Models Using Goodput	2024/06
Scaling Speculative Decoding with Lookahead Reasoning	2025/06

📖 Guided Decoding

Title	Date	arXiv	GitHub	Note	Read
Robust Text-to-SQL Generation with Execution-Guided Decoding	2018/07				✅
Efficient Guided Generation for Large Language Models	2023/07		Outlines		✅ ✅
XGrammar: Flexible and Efficient Structured Generation Engine for Large Language Models	2024/11		XGrammar
Pre3: Enabling Deterministic Pushdown Automata for Faster Structured LLM Generation	2025/06

📖 Long Sequence Processing

Title	Date	arXiv	GitHub	Note	Read

📖 Memory Offloading

Title	Date	arXiv	GitHub	Note	Read
MoE-Infinity: Efficient MoE Inference on Personal Machines with Sparsity-Aware Expert Cache	2024/01
ProMoE: Fast MoE-based LLM Serving using Proactive Caching	2024/10				✅

📖 Large Scale Serving

Title	Date	arXiv	GitHub	Note	Read
AccelGen: Heterogeneous SLO-Guaranteed High-Throughput LLM Inference Serving for Diverse Applications	2025/03
Serving Large Language Models on Huawei CloudMatrix384	2025/06

📖 Load Balancing

Title	Date	arXiv	GitHub	Note	Read
Lynx: Enabling Efficient MoE Inference through Dynamic Batch-Aware Expert Selection	2024/11
Pro-Prophet: A Systematic Load Balancing Method for Efficient Parallel Training of Large-scale MoE Models	2024/11

📖 KVCache Store

Title	Date	arXiv	GitHub	Note	Read
CacheBlend: Fast Large Language Model Serving for RAG with Cached Knowledge Fusion	2024/05		LMCache
Mooncake: A KVCache-centric Disaggregated Architecture for LLM Serving	2024/07		Mooncake

📖 Disaggregated Architecture

Title	Date	arXiv	GitHub	Note	Read
Splitwise: Efficient generative LLM inference using phase splitting	2023/11		splitwise-sim	link	✅
DistServe: Disaggregating Prefill and Decoding for Goodput-optimized Large Language Model Serving	2024/01		DistServe
MegaScale-Infer: Serving Mixture-of-Experts at Scale with Disaggregated Expert Parallelism	2025/04			link	✅ ✅
Step-3 is Large yet Affordable: Model-system Co-design for Cost-effective Decoding	2025/07		Step3, StepMesh	link	✅ ✅ ✅
xDeepServe: Model-as-a-Service on Huawei CloudMatrix384	2025/08

📖 Elasticity and Fault Tolerance

Title	Date	arXiv	GitHub	Note	Read
ServerlessLLM: Low-Latency Serverless Inference for Large Language Models	2024/01		ServerlessLLM	link	✅ ✅
Expert-as-a-Service: Towards Efficient, Scalable, and Robust Large-scale MoE Serving	2025/09			link	✅ ✅
ElasWave: An Elastic-Native System for Scalable Hybrid-Parallel Training	2025/10			link	✅ ✅
MoE-Prism: Disentangling Monolithic Experts for Elastic MoE Services via Model-System Co-Designs	2025/10			link	✅ ✅
From Models to Operators: Rethinking Autoscaling Granularity for Large Generative Models	2025/11			link	✅

📚 Learning Projects

Project	Category	Author/Organization	About
llm-action	LLM	@liguodongiot	本项目旨在分享大模型相关技术原理以及实战经验（大模型工程化、大模型应用落地）。
awesomeMLSys	MLSys	@GPU MODE	An ML Systems Onboarding list.
InfraTech	MLSys	@CalvinXKY	分享AI Infra知识&代码练习：PyTorch/vLLM/SGLang框架入门⚡️、性能加速🚀、大模型基础🧠、AI软硬件🔧等。
AI-Infra-from-Zero-to-Hero	MLSys	@HuaizhengZhang	🚀 Awesome System for Machine Learning ⚡️ AI System Papers and Industry Practice. ⚡️ System for Machine Learning, LLM (Large Language Model), GenAI (Generative AI). 🍻 OSDI, NSDI, SIGCOMM, SoCC, MLSys, etc. 🗃️ Llama3, Mistral, etc. 🧑‍💻 Video Tutorials.
resource-stream	CUDA	@GPU MODE	GPU programming related news and material links.
BasicCUDA	CUDA	@CalvinXKY	A tutorial for CUDA & PyTorch.

©️ Citation

@misc{cs-self-learning@2023,
  title  = {cs-self-learning},
  url    = {https://github.com/shen-shanshan/cs-self-learning},
  note   = {Open-source software available at https://github.com/shen-shanshan/cs-self-learning},
  author = {shen-shanshan},
  year   = {2023}
}

📜 License

MIT License, find more details here.

⭐ Star History