VectorStoreIndex: searching documents by meaning

A VectorStoreIndex embeds every document once and stores the vectors. A question is embedded and compared with all of them, the search lesson 2 did by hand.

Example

from llama_index.core import Settings
from llama_index.embeddings.huggingface import HuggingFaceEmbedding

Settings.embed_model = HuggingFaceEmbedding(model_name="sentence-transformers/all-MiniLM-L6-v2")
from llama_index.core import Document, VectorStoreIndex

documents = [
    Document(text="You can get a full refund within 30 days of delivery.", metadata={"file_name": "refunds.md"}),
    Document(text="Standard delivery takes 3 to 5 working days.", metadata={"file_name": "delivery.md"}),
    Document(text="The LMP-204 desk lamp has a known cable fault.", metadata={"file_name": "lamps.md"}),
]
index = VectorStoreIndex.from_documents(documents)
print(type(index).__name__, len(index.docstore.docs))

A Document is text plus metadata, a dictionary of facts about it such as the file it came from. from_documents splits each document into chunks, embeds them with Settings.embed_model and keeps them in memory.

Retrieving

Example

retriever = index.as_retriever(similarity_top_k=2)
for question in ["How do I get my money back?", "My parcel is late"]:
    results = retriever.retrieve(question)
    print(question, [(r.metadata["file_name"], round(r.score, 3)) for r in results])

as_retriever(similarity_top_k=2) returns the two closest chunks for a question, each with its score, the cosine similarity. Both questions that defeated keyword search in lesson 1 now find the right document first.

The docs call this top-k retrieval. It always returns k results, relevant or not: the second result for each question is simply the next closest document. Lesson 11 deals with results that are close enough to be returned and still irrelevant.

Try it yourself

Ask "LMP-204" and look at the scores.
Set similarity_top_k=3.
Add a fourth document about opening hours and ask when the shop is open.

PreviousEmbeddings: text as numbers that carry meaning

Slow is fine. Stopping is the only problem.