Skip to main content

Exercise 2 - Persist and Reload

Goal: Save your FAISS index to disk and reload it in a fresh process, without re-embedding the document.

Background​

In Exercise 1 you built the index every time the script ran. That means re-embedding the entire document on each run - slow and wasteful. In a real pipeline you build the index once and reload it instantly.

FAISS handles vector storage with faiss.write_index / faiss.read_index. But FAISS only stores vectors, not text. You need to save the chunk list separately.

Assignment​

Open 02_persist.py.

Part A - Save

  1. If index.faiss does not already exist, build the index from Exercise 1.
  2. Save the index: faiss.write_index(index, "index.faiss").
  3. Save the chunks list to chunks.json using the json module.
  4. Print the file sizes of both files.

Part B - Load

  1. Load the index: faiss.read_index("index.faiss").
  2. Load the chunks list from chunks.json.
  3. Run the same query as in Exercise 1 and confirm results are identical.
  4. Time the load. Compare it to the time it took to build from scratch.

Thinking questions​

  • You saved chunks as plain text in JSON. What metadata would you want to save alongside the text in a production system?
  • faiss.write_index only saves vectors and index structure - not the original Document objects or their page numbers. Design a simple file format (JSON or otherwise) that would store everything you need to fully reconstruct the search results.

← Exercise 1 · Next: Exercise 3 - ChromaDB →