Google Generative AI Embeddings (AI Studio & Gemini API)

Connect to Google's generative AI embeddings service using the GoogleGenerativeAIEmbeddings class, found in the langchain-google-genai package.

This will help you get started with Google's Generative AI embedding models (like Gemini) using LangChain. For detailed documentation on GoogleGenerativeAIEmbeddings features and configuration options, please refer to the API reference.

Overview

Integration details

Provider	Package
Google Gemini	langchain-google-genai

Setup

To access Google Generative AI embedding models you'll need to create a Google Cloud project, enable the Generative Language API, get an API key, and install the langchain-google-genai integration package.

Credentials

To use Google Generative AI models, you must have an API key. You can create one in Google AI Studio. See the Google documentation for instructions.

Once you have a key, set it as an environment variable GOOGLE_API_KEY:

import getpass
import os

if not os.getenv("GOOGLE_API_KEY"):
    os.environ["GOOGLE_API_KEY"] = getpass.getpass("Enter your Google API key: ")

To enable automated tracing of your model calls, set your LangSmith API key:

# os.environ["LANGSMITH_TRACING"] = "true"
# os.environ["LANGSMITH_API_KEY"] = getpass.getpass("Enter your LangSmith API key: ")

Installation

%pip install --upgrade --quiet  langchain-google-genai

Usage

from langchain_google_genai import GoogleGenerativeAIEmbeddings

embeddings = GoogleGenerativeAIEmbeddings(model="models/gemini-embedding-exp-03-07")
vector = embeddings.embed_query("hello, world!")
vector[:5]

API Reference:GoogleGenerativeAIEmbeddings

[-0.024917153641581535,
 0.012005362659692764,
 -0.003886754624545574,
 -0.05774897709488869,
 0.0020742062479257584]

Batch

You can also embed multiple strings at once for a processing speedup:

vectors = embeddings.embed_documents(
    [
        "Today is Monday",
        "Today is Tuesday",
        "Today is April Fools day",
    ]
)
len(vectors), len(vectors[0])

(3, 3072)

Indexing and Retrieval

Embedding models are often used in retrieval-augmented generation (RAG) flows, both as part of indexing data as well as later retrieving it. For more detailed instructions, please see our RAG tutorials.

Below, see how to index and retrieve data using the embeddings object we initialized above. In this example, we will index and retrieve a sample document in the InMemoryVectorStore.

# Create a vector store with a sample text
from langchain_core.vectorstores import InMemoryVectorStore

text = "LangChain is the framework for building context-aware reasoning applications"

vectorstore = InMemoryVectorStore.from_texts(
    [text],
    embedding=embeddings,
)

# Use the vectorstore as a retriever
retriever = vectorstore.as_retriever()

# Retrieve the most similar text
retrieved_documents = retriever.invoke("What is LangChain?")

# show the retrieved document's content
retrieved_documents[0].page_content

API Reference:InMemoryVectorStore

'LangChain is the framework for building context-aware reasoning applications'

Task type

GoogleGenerativeAIEmbeddings optionally support a task_type, which currently must be one of:

• SEMANTIC_SIMILARITY: Used to generate embeddings that are optimized to assess text similarity.
• CLASSIFICATION: Used to generate embeddings that are optimized to classify texts according to preset labels.
• CLUSTERING: Used to generate embeddings that are optimized to cluster texts based on their similarities.
• RETRIEVAL_DOCUMENT, RETRIEVAL_QUERY, QUESTION_ANSWERING, and FACT_VERIFICATION: Used to generate embeddings that are optimized for document search or information retrieval.
• CODE_RETRIEVAL_QUERY: Used to retrieve a code block based on a natural language query, such as sort an array or reverse a linked list. Embeddings of the code blocks are computed using RETRIEVAL_DOCUMENT.

By default, we use RETRIEVAL_DOCUMENT in the embed_documents method and RETRIEVAL_QUERY in the embed_query method. If you provide a task type, we will use that for all methods.

%pip install --upgrade --quiet  matplotlib scikit-learn

from langchain_google_genai import GoogleGenerativeAIEmbeddings
from sklearn.metrics.pairwise import cosine_similarity

query_embeddings = GoogleGenerativeAIEmbeddings(
    model="models/gemini-embedding-exp-03-07", task_type="RETRIEVAL_QUERY"
)
doc_embeddings = GoogleGenerativeAIEmbeddings(
    model="models/gemini-embedding-exp-03-07", task_type="RETRIEVAL_DOCUMENT"
)

q_embed = query_embeddings.embed_query("What is the capital of France?")
d_embed = doc_embeddings.embed_documents(
    ["The capital of France is Paris.", "Philipp is likes to eat pizza."]
)

for i, d in enumerate(d_embed):
    print(f"Document {i+1}:")
    print(f"Cosine similarity with query: {cosine_similarity([q_embed], [d])[0][0]}")
    print("---")

API Reference:GoogleGenerativeAIEmbeddings

Document 1
Cosine similarity with query: 0.7892893360164779
---
Document 2
Cosine similarity with query: 0.5438283285204146
---

API Reference

For detailed documentation on GoogleGenerativeAIEmbeddings features and configuration options, please refer to the API reference.

Additional Configuration

You can pass the following parameters to ChatGoogleGenerativeAI in order to customize the SDK's behavior:

• client_options: Client Options to pass to the Google API Client, such as a custom client_options["api_endpoint"]
• transport: The transport method to use, such as rest, grpc, or grpc_asyncio.

Related

• Embedding model conceptual guide
• Embedding model how-to guides