LangChain

Интеграция YDB с LangChain позволяет использовать YDB в качестве векторного хранилища для RAG приложений.

Эта интеграция позволяет разработчикам эффективно управлять, запрашивать и извлекать векторизованные данные, что является основой для современных приложений, связанных с обработкой естественного языка, поиском и анализом данных. Используя модели эмбеддингов, пользователи могут создавать сложные системы, способные понимать и извлекать информацию на основе семантического сходства.

Интеграция доступна для Python и JavaScript.

Установка

Для использования этой интеграции установите локальный YDB. Для получения дополнительной информации см. Установите и запустите YDB.

Также установите пакеты LangChain и модель эмбеддингов для нужного языка:

Python

JavaScript

pip install -qU langchain-ydb
pip install -qU langchain-huggingface

npm install @ydbjs/langchain @langchain/core
npm install @langchain/community @huggingface/transformers

Инициализация

Для создания векторного хранилища YDB требуется указать модель эмбеддингов и параметры подключения:

Python

JavaScript

from langchain_huggingface import HuggingFaceEmbeddings
from langchain_ydb.vectorstores import YDB, YDBSearchStrategy, YDBSettings

embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-mpnet-base-v2")

settings = YDBSettings(
    host="localhost",
    port=2136,
    database="/local",
    table="ydb_example",
    strategy=YDBSearchStrategy.COSINE_SIMILARITY,
)
vector_store = YDB(embeddings, config=settings)

import { HuggingFaceTransformersEmbeddings } from "@langchain/community/embeddings/huggingface_transformers";
import { YDBSearchStrategy, YDBVectorStore } from "@ydbjs/langchain";

const embeddings = new HuggingFaceTransformersEmbeddings({
  model: "sentence-transformers/all-mpnet-base-v2",
});

const vectorStore = new YDBVectorStore(embeddings, {
  connectionString: "grpc://localhost:2136/local",
  table: "ydb_example",
  strategy: YDBSearchStrategy.CosineSimilarity,
});

Управление векторным хранилищем

Создав векторное хранилище, можно взаимодействовать с ним, добавляя и удаляя различные элементы.

Добавление элементов

Подготовьте документы для работы:

Python

JavaScript

from uuid import uuid4

from langchain_core.documents import Document

uuids = [str(uuid4()) for _ in range(10)]
documents = [
    Document(
        page_content="I had chocolate chip pancakes and scrambled eggs for breakfast this morning.",
        metadata={"source": "tweet"},
        id=uuids[0],
    ),
    Document(
        page_content="The weather forecast for tomorrow is cloudy and overcast, with a high of 62 degrees.",
        metadata={"source": "news"},
        id=uuids[1],
    ),
    Document(
        page_content="Building an exciting new project with LangChain - come check it out!",
        metadata={"source": "tweet"},
        id=uuids[2],
    ),
    Document(
        page_content="Robbers broke into the city bank and stole $1 million in cash.",
        metadata={"source": "news"},
        id=uuids[3],
    ),
    Document(
        page_content="Wow! That was an amazing movie. I can't wait to see it again.",
        metadata={"source": "tweet"},
        id=uuids[4],
    ),
    Document(
        page_content="Is the new iPhone worth the price? Read this review to find out.",
        metadata={"source": "website"},
        id=uuids[5],
    ),
    Document(
        page_content="The top 10 soccer players in the world right now.",
        metadata={"source": "website"},
        id=uuids[6],
    ),
    Document(
        page_content="LangGraph is the best framework for building stateful, agentic applications!",
        metadata={"source": "tweet"},
        id=uuids[7],
    ),
    Document(
        page_content="The stock market is down 500 points today due to fears of a recession.",
        metadata={"source": "news"},
        id=uuids[8],
    ),
    Document(
        page_content="I have a bad feeling I am going to get deleted :(",
        metadata={"source": "tweet"},
        id=uuids[9],
    ),
]

Добавьте документы в векторное хранилище:

ids = vector_store.add_documents(documents=documents)

import { Document } from "@langchain/core/documents";
import { randomUUID } from "node:crypto";

const uuids = Array.from({ length: 10 }, () => randomUUID());
const documents = [
  new Document({
    pageContent: "I had chocolate chip pancakes and scrambled eggs for breakfast this morning.",
    metadata: { source: "tweet" },
    id: uuids[0],
  }),
  new Document({
    pageContent: "The weather forecast for tomorrow is cloudy and overcast, with a high of 62 degrees.",
    metadata: { source: "news" },
    id: uuids[1],
  }),
  new Document({
    pageContent: "Building an exciting new project with LangChain - come check it out!",
    metadata: { source: "tweet" },
    id: uuids[2],
  }),
  new Document({
    pageContent: "Robbers broke into the city bank and stole $1 million in cash.",
    metadata: { source: "news" },
    id: uuids[3],
  }),
  new Document({
    pageContent: "Wow! That was an amazing movie. I can't wait to see it again.",
    metadata: { source: "tweet" },
    id: uuids[4],
  }),
  new Document({
    pageContent: "Is the new iPhone worth the price? Read this review to find out.",
    metadata: { source: "website" },
    id: uuids[5],
  }),
  new Document({
    pageContent: "The top 10 soccer players in the world right now.",
    metadata: { source: "website" },
    id: uuids[6],
  }),
  new Document({
    pageContent: "LangGraph is the best framework for building stateful, agentic applications!",
    metadata: { source: "tweet" },
    id: uuids[7],
  }),
  new Document({
    pageContent: "The stock market is down 500 points today due to fears of a recession.",
    metadata: { source: "news" },
    id: uuids[8],
  }),
  new Document({
    pageContent: "I have a bad feeling I am going to get deleted :(",
    metadata: { source: "tweet" },
    id: uuids[9],
  }),
];

Добавьте документы в векторное хранилище:

const ids = await vectorStore.addDocuments(documents);

Удаление элементов

Элементы удаляются из векторного хранилища по идентификатору с использованием функции delete:

Python

JavaScript

vector_store.delete(ids=[ids[-1]])

await vectorStore.delete({ ids: [ids.at(-1)] });

Запросы в векторное хранилище

После создания векторного хранилища и добавления в него необходимых документов появляется возможность выполнения поисковых запросов в процессе выполнения цепочки или агента.

Прямой запрос

Поиск по сходству

Простой поиск по сходству можно выполнить следующим образом:

Python

JavaScript

results = vector_store.similarity_search(
    "LangChain provides abstractions to make working with LLMs easy",
    k=2,
)
for res in results:
    print(f"* {res.page_content} [{res.metadata}]")

const results = await vectorStore.similaritySearch(
  "LangChain provides abstractions to make working with LLMs easy",
  2,
);
for (const res of results) {
  console.log(`* ${res.pageContent} [${JSON.stringify(res.metadata)}]`);
}

Результат:

* Building an exciting new project with LangChain - come check it out! [{'source': 'tweet'}]
* LangGraph is the best framework for building stateful, agentic applications! [{'source': 'tweet'}]

Поиск по сходству с оценкой

Также можно выполнить поиск с оценкой:

Python

JavaScript

results = vector_store.similarity_search_with_score(
    "Will it be hot tomorrow?",
    k=3,
)
for res, score in results:
    print(f"* [SIM={score:.3f}] {res.page_content} [{res.metadata}]")

const results = await vectorStore.similaritySearchWithScore(
  "Will it be hot tomorrow?",
  3,
);
for (const [res, score] of results) {
  console.log(`* [SIM=${score.toFixed(3)}] ${res.pageContent} [${JSON.stringify(res.metadata)}]`);
}

Результат:

* [SIM=0.595] The weather forecast for tomorrow is cloudy and overcast, with a high of 62 degrees. [{'source': 'news'}]
* [SIM=0.212] I had chocolate chip pancakes and scrambled eggs for breakfast this morning. [{'source': 'tweet'}]
* [SIM=0.118] Wow! That was an amazing movie. I can't wait to see it again. [{'source': 'tweet'}]

Фильтрация

Поиск с использованием фильтров выполняется следующим образом:

Python

JavaScript

results = vector_store.similarity_search_with_score(
    "What did I eat for breakfast?",
    k=4,
    filter={"source": "tweet"},
)
for res, _ in results:
    print(f"* {res.page_content} [{res.metadata}]")

const results = await vectorStore.similaritySearchWithScore(
  "What did I eat for breakfast?",
  4,
  { source: "tweet" },
);
for (const [res] of results) {
  console.log(`* ${res.pageContent} [${JSON.stringify(res.metadata)}]`);
}

Результат:

* I had chocolate chip pancakes and scrambled eggs for breakfast this morning. [{'source': 'tweet'}]
* Wow! That was an amazing movie. I can't wait to see it again. [{'source': 'tweet'}]
* Building an exciting new project with LangChain - come check it out! [{'source': 'tweet'}]
* LangGraph is the best framework for building stateful, agentic applications! [{'source': 'tweet'}]

Запрос через трансформацию в ретривер

Векторное хранилище можно трансформировать в поисковик (retriever) для упрощённого использования в цепочках.

Пример представлен ниже:

Python

JavaScript

retriever = vector_store.as_retriever(
    search_kwargs={
        "k": 2,
        "filter": {"source": "news"},
    },
)
results = retriever.invoke("Stealing from the bank is a crime")
for res in results:
    print(f"* {res.page_content} [{res.metadata}]")

const retriever = vectorStore.asRetriever({
  k: 2,
  filter: { source: "news" },
});
const results = await retriever.invoke("Stealing from the bank is a crime");
for (const res of results) {
  console.log(`* ${res.pageContent} [${JSON.stringify(res.metadata)}]`);
}

Результат:

* Robbers broke into the city bank and stole $1 million in cash. [{'source': 'news'}]
* The stock market is down 500 points today due to fears of a recession. [{'source': 'news'}]

Была ли статья полезна?

Векторный поиск

YDB CLI