Course Wiki

Embeddings

Representing text as vectors so similarity in meaning becomes closeness in space.

An embedding model maps text to a fixed-length vector such that texts with similar meanings land close together. LLM Zoomcamp’s module 2 embeds the FAQ documents and the user’s question into the same vector space, and retrieval becomes nearest-neighbor search: the documents whose vectors are closest to the question vector are the relevant ones.

The course covers the operational realities alongside the math: choosing an embedding model, computing and storing vectors (with sqlitesearch, minsearch, and PGVector as the course’s storage options), and the cost model of calling an embedding API over a whole knowledge base.

Taught in

• LLM Zoomcamp

  • Module 2: Vector Search
    • Vector Search
    • Embeddings
    • Embedding Our Dataset
    • Vector Search
    • Vector Search with minsearch
    • RAG with Vector Search
    • Vector Search with sqlitesearch
    • Vector Search with PGVector
    • Using ONNX Runtime instead of PyTorch
    • Next Steps
  • Module 3: Orchestration
    • Retrieval Augmented Generation
  • Module 6: Best Practices
    • Hybrid Search with LangChain
    • Next Steps
  • Module 7: End-to-End Project
    • Chunking for Longer Texts ## Related concepts

• Vector Search

• RAG

• Keyword Search

• Hybrid Search