From 0824d65a5c73f3b51e9f1c9910851197211fb44b Mon Sep 17 00:00:00 2001 From: Harrison Chase Date: Mon, 27 Feb 2023 00:31:36 -0800 Subject: [PATCH] Harrison/indexing pipeline (#1317) --- .../chain_examples/question_answering.ipynb | 23 +-- docs/modules/indexes/getting_started.ipynb | 184 +++++++++++++++--- langchain/indexes/__init__.py | 3 +- langchain/indexes/vectorstore.py | 37 ++++ 4 files changed, 211 insertions(+), 36 deletions(-) create mode 100644 langchain/indexes/vectorstore.py diff --git a/docs/modules/indexes/chain_examples/question_answering.ipynb b/docs/modules/indexes/chain_examples/question_answering.ipynb index d6cbc764e3..1a1981392e 100644 --- a/docs/modules/indexes/chain_examples/question_answering.ipynb +++ b/docs/modules/indexes/chain_examples/question_answering.ipynb @@ -30,29 +30,24 @@ "from langchain.text_splitter import CharacterTextSplitter\n", "from langchain.vectorstores import Chroma\n", "from langchain.docstore.document import Document\n", - "from langchain.prompts import PromptTemplate" + "from langchain.prompts import PromptTemplate\n", + "from langchain.indexes.vectorstore import VectorstoreIndexCreator" ] }, { "cell_type": "code", "execution_count": 2, - "id": "291f0117", + "id": "ef9305cc", "metadata": {}, "outputs": [], "source": [ - "from langchain.document_loaders import TextLoader\n", - "loader = TextLoader('../../state_of_the_union.txt')\n", - "documents = loader.load()\n", - "text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)\n", - "texts = text_splitter.split_documents(documents)\n", - "\n", - "embeddings = OpenAIEmbeddings()" + "index_creator = VectorstoreIndexCreator()" ] }, { "cell_type": "code", - "execution_count": 4, - "id": "fd9666a9", + "execution_count": 3, + "id": "291f0117", "metadata": {}, "outputs": [ { @@ -65,12 +60,14 @@ } ], "source": [ - "docsearch = Chroma.from_documents(texts, embeddings)" + "from langchain.document_loaders import TextLoader\n", + "loader = TextLoader('../../state_of_the_union.txt')\n", + "docsearch = index_creator.from_loaders([loader])" ] }, { "cell_type": "code", - "execution_count": 5, + "execution_count": 4, "id": "d1eaf6e6", "metadata": {}, "outputs": [], diff --git a/docs/modules/indexes/getting_started.ipynb b/docs/modules/indexes/getting_started.ipynb index cbe047a1b9..4a2678eea7 100644 --- a/docs/modules/indexes/getting_started.ipynb +++ b/docs/modules/indexes/getting_started.ipynb @@ -2,45 +2,152 @@ "cells": [ { "cell_type": "markdown", - "id": "07c1e3b9", + "id": "2244801b", "metadata": {}, "source": [ "# Getting Started\n", "\n", - "This example showcases question answering over a vector database.\n", - "We have chosen this as the example for getting started because it nicely combines a lot of different elements (Text splitters, embeddings, vectorstores) and then also shows how to use them in a chain." + "This example showcases question answering over documents.\n", + "We have chosen this as the example for getting started because it nicely combines a lot of different elements (Text splitters, embeddings, vectorstores) and then also shows how to use them in a chain.\n", + "\n", + "Question answering over documents consists of three steps:\n", + "\n", + "1. Create an index\n", + "2. Create a question answering chain\n", + "3. Ask questions!\n", + "\n", + "Each of the steps has multiple sub steps and potential configurations. In this notebook we will primarily focus on (1). We will start by showing the one-liner for doing so, but then break down what is actually going on.\n", + "\n", + "First, let's import some common classes we'll use no matter what." ] }, { "cell_type": "code", "execution_count": 1, - "id": "82525493", + "id": "8d369452", "metadata": {}, "outputs": [], "source": [ - "from langchain.embeddings.openai import OpenAIEmbeddings\n", - "from langchain.vectorstores import Chroma\n", - "from langchain.text_splitter import CharacterTextSplitter\n", - "from langchain import OpenAI, VectorDBQA" + "from langchain.chains import VectorDBQA\n", + "from langchain.llms import OpenAI" ] }, { "cell_type": "markdown", - "id": "0b7adc54", + "id": "07c1e3b9", "metadata": {}, "source": [ - "Here we load in the documents we want to use to create our index." + "Next in the generic setup, let's specify the document loader we want to use." + ] + }, + { + "cell_type": "code", + "execution_count": 2, + "id": "33958a86", + "metadata": {}, + "outputs": [], + "source": [ + "from langchain.document_loaders import TextLoader\n", + "loader = TextLoader('../state_of_the_union.txt')" + ] + }, + { + "cell_type": "markdown", + "id": "489c74bb", + "metadata": {}, + "source": [ + "## One Line Index Creation\n", + "\n", + "To get started as quickly as possible, we can use the `VectorstoreIndexCreator`." ] }, { "cell_type": "code", "execution_count": 3, - "id": "611e0c19", + "id": "403fc231", + "metadata": {}, + "outputs": [], + "source": [ + "from langchain.indexes import VectorstoreIndexCreator" + ] + }, + { + "cell_type": "code", + "execution_count": 4, + "id": "57a8a199", + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Running Chroma using direct local API.\n", + "Using DuckDB in-memory for database. Data will be transient.\n" + ] + } + ], + "source": [ + "index = VectorstoreIndexCreator().from_loaders([loader])" + ] + }, + { + "cell_type": "markdown", + "id": "f3493fa4", + "metadata": {}, + "source": [ + "Now that the index is created, we can use it in a VectorDBQAChain to ask questions of the data!" + ] + }, + { + "cell_type": "code", + "execution_count": 5, + "id": "23d0d234", + "metadata": {}, + "outputs": [ + { + "data": { + "text/plain": [ + "\" The president said that Ketanji Brown Jackson is one of the nation's top legal minds, a consensus builder, and has gained a broad range of support. He also said that she is a former top litigator in private practice, a former federal public defender, and from a family of public school educators and police officers.\"" + ] + }, + "execution_count": 5, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "qa = VectorDBQA.from_chain_type(llm=OpenAI(), chain_type=\"stuff\", vectorstore=index)\n", + "query = \"What did the president say about Ketanji Brown Jackson\"\n", + "qa.run(query)" + ] + }, + { + "cell_type": "markdown", + "id": "2cb6d2eb", + "metadata": {}, + "source": [ + "## Walkthrough\n", + "\n", + "Okay, so what's actually going on? How is this index getting created?\n", + "\n", + "A lot of the magic is being hid in this `VectorstoreIndexCreator`. What is this doing?\n", + "\n", + "There are three main steps going on after the documents are loaded:\n", + "\n", + "1. Splitting documents into chunks\n", + "2. Creating embeddings for each document\n", + "3. Storing documents and embeddings in a vectorstore\n", + "\n", + "Let's walk through this in code" + ] + }, + { + "cell_type": "code", + "execution_count": 6, + "id": "54270abc", "metadata": {}, "outputs": [], "source": [ - "from langchain.document_loaders import TextLoader\n", - "loader = TextLoader('../state_of_the_union.txt')\n", "documents = loader.load()" ] }, @@ -54,11 +161,12 @@ }, { "cell_type": "code", - "execution_count": 4, + "execution_count": 8, "id": "afecb8cf", "metadata": {}, "outputs": [], "source": [ + "from langchain.text_splitter import CharacterTextSplitter\n", "text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)\n", "texts = text_splitter.split_documents(documents)" ] @@ -73,11 +181,12 @@ }, { "cell_type": "code", - "execution_count": 5, + "execution_count": 10, "id": "9eaaa735", "metadata": {}, "outputs": [], "source": [ + "from langchain.embeddings import OpenAIEmbeddings\n", "embeddings = OpenAIEmbeddings()" ] }, @@ -91,7 +200,7 @@ }, { "cell_type": "code", - "execution_count": 6, + "execution_count": 11, "id": "5c7049db", "metadata": {}, "outputs": [ @@ -105,6 +214,7 @@ } ], "source": [ + "from langchain.vectorstores import Chroma\n", "db = Chroma.from_documents(texts, embeddings)" ] }, @@ -113,12 +223,12 @@ "id": "30c4e5c6", "metadata": {}, "source": [ - "Finally, we create a chain and use it to answer questions!" + "Then, as before, we create a chain and use it to answer questions!" ] }, { "cell_type": "code", - "execution_count": 9, + "execution_count": 12, "id": "3018f865", "metadata": {}, "outputs": [], @@ -128,17 +238,17 @@ }, { "cell_type": "code", - "execution_count": 10, + "execution_count": 13, "id": "032a47f8", "metadata": {}, "outputs": [ { "data": { "text/plain": [ - "\" The President said that Ketanji Brown Jackson is one of the nation's top legal minds, a former top litigator in private practice, a former federal public defender, and from a family of public school educators and police officers. He said that she is a consensus builder and has received a broad range of support from the Fraternal Order of Police to former judges appointed by Democrats and Republicans.\"" + "\" The President said that Ketanji Brown Jackson is one of the nation's top legal minds and a consensus builder, with a broad range of support from the Fraternal Order of Police to former judges appointed by Democrats and Republicans. She is a former top litigator in private practice, a former federal public defender, and from a family of public school educators and police officers.\"" ] }, - "execution_count": 10, + "execution_count": 13, "metadata": {}, "output_type": "execute_result" } @@ -148,10 +258,40 @@ "qa.run(query)" ] }, + { + "cell_type": "markdown", + "id": "9464690e", + "metadata": {}, + "source": [ + "`VectorstoreIndexCreator` is just a wrapper around all this logic. It is configurable in the text splitter it uses, the embeddings it uses, and the vectorstore it uses. For example, you can configure it as below:" + ] + }, + { + "cell_type": "code", + "execution_count": 14, + "id": "4001bbc6", + "metadata": {}, + "outputs": [], + "source": [ + "index_creator = VectorstoreIndexCreator(\n", + " vectorstore_cls=Chroma, \n", + " embedding=OpenAIEmbeddings(),\n", + " text_splitter=CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)\n", + ")" + ] + }, + { + "cell_type": "markdown", + "id": "78d8d143", + "metadata": {}, + "source": [ + "Hopefully this highlights what is going on under the hood of `VectorstoreIndexCreator`. While we think it's important to have a simple way to create indexes, we also think it's important to understand what's going on under the hood." + ] + }, { "cell_type": "code", "execution_count": null, - "id": "8b403637", + "id": "dd7257bf", "metadata": {}, "outputs": [], "source": [] diff --git a/langchain/indexes/__init__.py b/langchain/indexes/__init__.py index c8a3547bfe..7e81e4ac88 100644 --- a/langchain/indexes/__init__.py +++ b/langchain/indexes/__init__.py @@ -1,4 +1,5 @@ """All index utils.""" from langchain.indexes.graph import GraphIndexCreator +from langchain.indexes.vectorstore import VectorstoreIndexCreator -__all__ = ["GraphIndexCreator"] +__all__ = ["GraphIndexCreator", "VectorstoreIndexCreator"] diff --git a/langchain/indexes/vectorstore.py b/langchain/indexes/vectorstore.py new file mode 100644 index 0000000000..f277247ffa --- /dev/null +++ b/langchain/indexes/vectorstore.py @@ -0,0 +1,37 @@ +from typing import List, Type + +from pydantic import BaseModel, Extra, Field + +from langchain.document_loaders.base import BaseLoader +from langchain.embeddings.base import Embeddings +from langchain.embeddings.openai import OpenAIEmbeddings +from langchain.text_splitter import RecursiveCharacterTextSplitter, TextSplitter +from langchain.vectorstores.base import VectorStore +from langchain.vectorstores.chroma import Chroma + + +def _get_default_text_splitter() -> TextSplitter: + return RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=0) + + +class VectorstoreIndexCreator(BaseModel): + """Logic for creating indexes.""" + + vectorstore_cls: Type[VectorStore] = Chroma + embedding: Embeddings = Field(default_factory=OpenAIEmbeddings) + text_splitter: TextSplitter = Field(default_factory=_get_default_text_splitter) + + class Config: + """Configuration for this pydantic object.""" + + extra = Extra.forbid + arbitrary_types_allowed = True + + def from_loaders(self, loaders: List[BaseLoader]) -> VectorStore: + """Create a vectorstore index from loaders.""" + docs = [] + for loader in loaders: + docs.extend(loader.load()) + sub_docs = self.text_splitter.split_documents(docs) + vectorstore = self.vectorstore_cls.from_documents(sub_docs, self.embedding) + return vectorstore