Hands-on LLMs

This learning path focuses on practical implementation, architecture, and deployment of Large Language Models, following the textbook Hands-On Large Language Models by Jay Alammar and Maarten Grootendorst.

📅 Progress & Chapter Summaries

Chapter 1: Introduction to Language Models

  • Summary: High-level overview of language models, history, and applications (translation, summarization, chatbots, classification).
  • Core Concepts:
    • Loading models using the Hugging Face pipeline abstraction.
    • Interfacing with local pre-trained weights (e.g., microsoft/Phi-3-mini-4k-instruct).
    • Configuring GPUs (CUDA) for model inference.
  • Concept Links:
    • [[large-language-models]]

Chapter 2: Tokens and Token Embeddings

  • Summary: Deep-dive into how text is converted into numbers and subsequently mapped to continuous vector spaces.
  • Core Concepts:
    • Tokenization: Breaking words into subword pieces using BPE, WordPiece, or SentencePiece. Comparing tokenizers across model generations (e.g., GPT’s tiktoken vs. Llama vs. Phi-3).
    • Word Embeddings: Static embeddings (Word2Vec) vs. contextualized embeddings (BERT, GPT, Phi-3).
    • Sentence Embeddings: Computing semantic similarity between entire sentences using pooling mechanisms and sentence-transformers.
    • Beyond Text: Using item sequence embeddings (e.g., songs in user sessions) to train custom recommendation algorithms with gensim.
  • Concept Links:
    • [[tokenization]]
    • [[embeddings]]

🛠️ Resources & Environment Setup