Course Wiki

Embedding Our Dataset — LLM Zoomcamp Module 2

--- video_url: 'https://www.youtube.com/watch?v=NC89mz1iG4E&list=PL3MmuxUbc_hLZFNgSad56pDBKK8KO0XIv' --- # Embedding Our Dataset

LLM Zoomcamp › Module 2: Vector Search › Embedding Our Dataset

Notes

video_url: “https://www.youtube.com/watch?v=NC89mz1iG4E&list=PL3MmuxUbc_hLZFNgSad56pDBKK8KO0XIv”

Embedding Our Dataset

In the previous lesson we saw how embeddings work on a couple of examples. Now we apply them to the whole FAQ dataset.

Loading the data

In module 1 we created ingest.py for loading the FAQ data.

Download it into your project:

wget https://raw.githubusercontent.com/DataTalksClub/llm-zoomcamp/main/cohorts/2026/01-agentic-rag/code/ingest.py

We use it here:

from ingest import load_faq_data
documents = load_faq_data()

Generating embeddings

Each document is a Python dictionary with a question and an answer. We embed both together. That way a query can match against the question text and the answer text in our index.

Build one text per document:

texts = []
for doc in documents:
    text = doc["question"] + " " + doc["answer"]
    texts.append(text)

Now we generate the embeddings. We have about 1200 texts here. We won’t hand the model all of them at once. That takes a long time, and we can’t see what’s happening inside. Instead we split them into batches.

First we import tqdm so we can watch the progress:

from tqdm.auto import tqdm

Next we chunk the dataset into batches of 50 and encode each batch:

batch_size = 50
vectors = []
for i in tqdm(range(0, len(texts), batch_size)):
    batch = texts[i:i + batch_size]
    batch_vectors = model.encode(batch)
    vectors.extend(batch_vectors)
len(vectors)

We end up with 1208 vectors. On a GPU this is fast. Most of us run on Codespaces without a GPU, so it takes a bit, but it’s a one-off.

We turn them into a 2-dimensional array (matrix) where

• rows are documents (vectors)
• columns are dimensions of the vectors

import numpy as np
X = np.array(vectors)

Calling X.shape returns (1208, 384) - number of documents vs number of dimensions.

Key concepts

• Embeddings

Related notes

• llmz-m02-embeddings
• llmz-m02-next-steps
• llmz-m02-rag-with-vector-search
• llmz-m02-using-onnx-runtime-instead-of-pytorch
• llmz-m02-vector-search
• llmz-m02-vector-search-2
• llmz-m02-vector-search-with-minsearch
• llmz-m02-vector-search-with-pgvector
• llmz-m02-vector-search-with-sqlitesearch
• llmz-m03-retrieval-augmented-generation
• llmz-m06-hybrid-search-with-langchain
• llmz-m06-next-steps
• llmz-m07-chunking-for-longer-texts

Sources

• Video
• Lesson file