Add notebooks for multi-vector search course

course-multi-vector-search/module-0/installing-dependencies.ipynb

@@ -0,0 +1,111 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "id": "9af5bc4b49d17ebc",
+   "metadata": {},
+   "source": [
+    "# Module 0: Installing Dependencies"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "mli6s9xkllr",
+   "metadata": {},
+   "source": [
+    "We need `qdrant-client` for interacting with Qdrant and `fastembed` for generating vector embeddings locally."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "466eb04e8e447feb",
+   "metadata": {
+    "jupyter": {
+     "is_executing": true
+    }
+   },
+   "outputs": [],
+   "source": [
+    "!pip install -q \"qdrant-client>=1.16.2\" \"fastembed>=0.8.0\""
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "6u2wb574s8n",
+   "metadata": {},
+   "source": [
+    "Let's verify that both packages were installed correctly by importing them."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "8e66867cd6ebf200",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from qdrant_client import QdrantClient\n",
+    "from fastembed import TextEmbedding\n",
+    "\n",
+    "print(\"All dependencies installed successfully!\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "uvxeas5v8er",
+   "metadata": {},
+   "source": [
+    "Connect to your Qdrant instance. Replace the URL and API key with your own Qdrant Cloud credentials, or use a local instance."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "1ea6882564a624f5",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# For Qdrant Cloud\n",
+    "client = QdrantClient(\n",
+    "    url=\"https://your-cluster-url.cloud.qdrant.io\",\n",
+    "    api_key=\"your-api-key\"\n",
+    ")\n",
+    "\n",
+    "# For local Qdrant\n",
+    "# client = QdrantClient(url=\"http://localhost:6333\")\n",
+    "\n",
+    "print(f\"Connected to Qdrant: {client.get_collections()}\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "a9bdeae2-ae97-4110-92f4-531168f3b31c",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.13.5"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

course-multi-vector-search/module-1/late-interaction-basics.ipynb

@@ -0,0 +1,214 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "id": "3ebe879db87a4432",
+   "metadata": {},
+   "source": [
+    "# Module 1: Late Interaction Basics"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "9b9p5pv2oh9",
+   "metadata": {},
+   "source": [
+    "Define a simple query and two documents to compare how different embedding approaches represent them."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "396986ed-a976-41a0-aad3-6de8f2406c5b",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "documents = [\n",
+    "    \"Qdrant is an AI-native vector database and a semantic search engine\",\n",
+    "    \"Relational databases are not well-suited for search\",\n",
+    "]\n",
+    "query = \"What is Qdrant?\""
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "11591ce2-f88c-438d-9594-beb4d7cff0bb",
+   "metadata": {},
+   "source": [
+    "## Single-Vector Embeddings (No Interaction)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "a54183c0-4a75-4981-a06c-3f1bc14b6b9b",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from fastembed import TextEmbedding\n",
+    "\n",
+    "# Load the BAAI/bge-small-en-v1.5 model\n",
+    "dense_model = TextEmbedding(\"BAAI/bge-small-en-v1.5\")\n",
+    "# Pass the documents through the model. The .passage_embed \n",
+    "# method returns a generator we can iterate over and is \n",
+    "# supposed to be used for the documents only.\n",
+    "dense_generator = dense_model.passage_embed(documents)\n",
+    "# Running next on the generator yields one vector at\n",
+    "# the time, representing a single document.\n",
+    "dense_vector = next(dense_generator)\n",
+    "dense_vector.shape"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "21f340ad-8ba3-4572-84ee-6171cdf5aa38",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Generate a dense vector for the query as well, using \n",
+    "# the .query_embed method this time.\n",
+    "dense_query_vector = next(dense_model.query_embed(query))\n",
+    "dense_query_vector.shape"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "wxe4h2vwnue",
+   "metadata": {},
+   "source": [
+    "Compute dot product similarity between the query and each document. Higher scores indicate stronger semantic matches."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "d7416f7b-c2d1-4642-b3b6-a5df3549a8b9",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import numpy as np\n",
+    "\n",
+    "# Calculate the dot product between the query\n",
+    "# and the first document vector\n",
+    "np.dot(dense_query_vector, dense_vector)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "2zgkv6yd2r4",
+   "metadata": {},
+   "source": [
+    "Now compare the same query against the second document."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "bada239c-dc12-47d5-8923-407452cb7731",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Calculate the dot product between the same query\n",
+    "# and the second document vectors\n",
+    "np.dot(dense_query_vector, next(dense_generator))"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "811caba5-69d2-45bc-9552-5aeac7adf37b",
+   "metadata": {},
+   "source": [
+    "## Cross-Encoders (Early Interaction)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "e995ba5b-ebd4-49e1-b06f-69f0cdd9052f",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from fastembed.rerank.cross_encoder import TextCrossEncoder\n",
+    "\n",
+    "# Load the Xenova/ms-marco-MiniLM-L-6-v2 cross encoder model\n",
+    "cross_encoder = TextCrossEncoder(\"Xenova/ms-marco-MiniLM-L-6-v2\")\n",
+    "# Run .rerank method on the query and all the documents.\n",
+    "# It does not create any vector representations, but gives\n",
+    "# the score indicating the relevance of the document for\n",
+    "# the provided query.\n",
+    "score_generator = cross_encoder.rerank(query, documents)\n",
+    "list(score_generator)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "9c154200-a65c-4f47-bc97-e0d092d2623f",
+   "metadata": {},
+   "source": [
+    "## Late Interaction: The Core Paradigm\n",
+    "\n",
+    "### The ColBERT Approach"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "27c39129-3acb-4876-ba07-5cda22f00bcc",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from fastembed import LateInteractionTextEmbedding\n",
+    "\n",
+    "# Load the colbert-ir/colbertv2.0 model\n",
+    "colbert_model = LateInteractionTextEmbedding(\"colbert-ir/colbertv2.0\")\n",
+    "# Run .passage_embed on all the documents and create\n",
+    "# a generator of the multi-vector representations\n",
+    "colbert_generator = colbert_model.passage_embed(documents)\n",
+    "colbert_vector = next(colbert_generator)\n",
+    "colbert_vector.shape"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "6f88562d-deef-4ce5-a448-0c854742521c",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Create multi-vector representation for the query\n",
+    "colbert_query_vector = next(colbert_model.query_embed(query))\n",
+    "colbert_query_vector.shape"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "a768b8df-75a0-4c66-a5d9-5c7be934af0a",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.13.5"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}