LLM Learning Initial

Welcome to the world of Large Language Models

For this summer vacation, I will begin the learning process for the fundamental principle and downstream techniques and applications for LLM (Large Language Models). I hope to find the journey interesting and fruitful!

All the code will be refactored and categorized into LLM Learn

LLM Learning Materials

• Courses and Videos

  • Post Training by DeepLearning Ai

  • CS25: Transformers United V5: For advanced architectures for transformers.

  • Transformers for 3b1b: A good visualize demo!

  • Advanced Natural Language Processing: A very good course for detailed lecture notes and homework. (I will try to finish that project in the next semester.)

  • LLMs and Transformers: with several discussion topics (lecture notes, blogs and papers are included)

  • Dive into LLMs: The chinese version of learning large language models

  • Learning Large Language Models from scratch: The course for learning LLMs in stanford.

  • Large Language Models for DataWhale: Courses of Chinese version.

• Several books and codes:

  • Hands on Large Language Models (English Version)

    • Original English Version: Hands on Large Language Models

    • https://www.llm-book.com/

  • Hands on Large Language Models (CHN Version)

    • Source Code

  • Build a Large Language Model (From Scratch)

    • Github Repo

    • Additional Technical Blog

• Projects:

  • LLM Hero to Zero:

    • Build a simple GPT from scratch.

    • LLM from hero to zero, karpathy version

      • Source Code

      • Lectures Videos

    • LLM from hero to zero, CHN version

      • Source Code

• Tool Usage

  • HuggingFace for downloading models and datasets

  • vllm

LLM Learning Contents

• Basic architecture for Large Language Models

  • Attention Mechanism (Attention is all you need!) ✅

  • RNN, LSTM, GRU (will be covered in the future)

  • Seq2Seq Model ✅

  • Transformer Architecture ✅

  • Other basic NLP knowledge (word embeddings, etc.)

• Pre Training for LLM

  • Loading Datasets

  • Self-supervised Learning

  • More advanced architecture for LLM pre-training, see advanced structure part.

• Post Training for LLM

  • Quantization for Model Optimization

  • Knowledge Distillation

  • Fine-tuning Techniques

    • SFT
    • RFT
    • RLHF (Reinforcement Learning from Human Feedback)

  • LLM Evaluation

• Advanced Structure for LLM

  • Advanced Transformer

  • Sparse Attention & Lightning Attention

  • KV cache

  • Mixture of Experts (MoE)

    • MoE Introduction

    • MoE Advanced

  • LoRA: Low-Rank Adaptation of Large Language Models

  • PPO, GRPO, DPO, etc. (Deep Reinforcement Learning)

• Test time compute for LLM (after training)

  • LLM Reasoning (CoT, ToT, etc.)

    • Recommend Blog: Why we think by Lilian Weng

• LLM DownStream Applications

  • This section will be recorded in the future.

  • RAG

  • LangChain Community

Updating Status

• 2025/07/28: Finish two long-standing blog posts: AINN-Attention & AINN-Transformer

  • Finish tutorial for basic Attention mechanism and Transformer Structure.

Current Todo List

• Finish the implementation code of Transformer Module in dl2ai.

• Learning courses: Word Embedding and basic NLP knowledge.