OpenAI

Let’s load the OpenAI Embedding class.

Setup

First we install langchain-openai and set the required env vars

%pip install -qU langchain-openai

import getpass
import os

os.environ["OPENAI_API_KEY"] = getpass.getpass()

from langchain_openai import OpenAIEmbeddings

API Reference:

• OpenAIEmbeddings

embeddings = OpenAIEmbeddings(model="text-embedding-3-large")

text = "This is a test document."

Usage

Embed query

query_result = embeddings.embed_query(text)

Warning: model not found. Using cl100k_base encoding.

query_result[:5]

[-0.014380056377383358,
 -0.027191711627651764,
 -0.020042716111860304,
 0.057301379620345545,
 -0.022267658631828974]

Embed documents

doc_result = embeddings.embed_documents([text])

Warning: model not found. Using cl100k_base encoding.

doc_result[0][:5]

[-0.014380056377383358,
 -0.027191711627651764,
 -0.020042716111860304,
 0.057301379620345545,
 -0.022267658631828974]

Specify dimensions

With the text-embedding-3 class of models, you can specify the size of the embeddings you want returned. For example by default text-embedding-3-large returned embeddings of dimension 3072:

len(doc_result[0])

3072

But by passing in dimensions=1024 we can reduce the size of our embeddings to 1024:

embeddings_1024 = OpenAIEmbeddings(model="text-embedding-3-large", dimensions=1024)

len(embeddings_1024.embed_documents([text])[0])

Warning: model not found. Using cl100k_base encoding.

1024