LLM Primer – Systematic LLM Learning

LLM Primer

Systematic Learning of Large Language Models: From Theory to Practice

A bilingual, multi-tiered, community-driven LLM learning knowledge base covering foundational concepts, engineering practices, cutting-edge papers, and interactive demos.

What / Who / How Long

• 30 minutes: Build mental models through intuitive explanations, understand core LLM concepts
• 3 hours: Complete engineering tier, master practical implementation and performance optimization
• On-demand deep dive: Track paper lineage, open questions, and frontier developments in research tier

Three-Tier Learning Mode

💡Intro Tier

Build intuitive understanding through analogies, visualizations, and interactive demos. Quickly establish mental models.

⚙️Engineer Tier

Focus on implementation details, performance optimization, cost trade-offs, and engineering best practices.

🔬Research Tier

Deep dive into paper details, theoretical assumptions, experimental evidence, and open questions. Connect to academic frontiers.

Four Modules & Learning Paths

🧠Foundations

Tokenization, Attention, Transformer, Sampling & Decoding, Positional Encoding, Embeddings, Emergent Abilities

7 articles · ~56 minutes⚡Training

Pretraining & Scaling Laws, Fine-Tuning & Alignment (SFT, RLHF, DPO)

4 articles · ~32 minutes🚀Inference

KV Cache & Quantization, FlashAttention & Efficient Attention, Long Context

3 articles · ~24 minutes🎯Applications

RAG & Retrieval Augmentation, Agents & Tool Use, Prompt Engineering

2 articles · ~16 minutes

Start Here

🌱 I’m a BeginnerStart with foundations module overview, follow recommended order⚙️ I’m an EngineerJump to inference & optimization module for practical solutions🔬 I’m a ResearcherBrowse paper library, explore classifications and citations🗺️ I Want the MapVisualize learning paths, get a bird’s-eye view

Recently Updated

• Cluster Networking & Communication: How GPUs Talk 2026-10-19

  NVLink and InfiniBand, RDMA and collective communication, scale-up vs scale-out, and the networking engineering of 10k-GPU clusters.

• Fine-Tuning and Alignment: Making Models Follow Instructions 2026-10-19

  SFT, RLHF, DPO, and preference optimization methods.

• Parallelism & Performance: Squeezing the GPUs 2026-10-19

  Data/tensor/pipeline parallelism, ZeRO and FSDP, activation recomputation, FlashAttention, MFU, and pipeline bubbles.

• Pretraining and Scaling Law: How Models Learn 2026-10-19

  Pretraining objectives, data engineering, scaling laws, and compute-optimal training.

• Weight Update & Optimizers: How Parameters Move 2026-10-19

  From AdamW to Muon: first/second moments, learning-rate schedules, mixed precision, gradient clipping, and training-stability engineering.

• Agents and Tool Use: Models Are More Than Chat 2026-09-13

  LLM agents, tool calling, planning, and multi-agent collaboration.

---

Subscribe RSS

Subscribe to blog updates, get new articles first