Amazon Bedrock

Select a language

Python

JavaScript

Chroma provides a convinient wrapper for Amazon Bedrock embedding API. This embedding function runs remotely on Amazon's servers, and requires an Amazon client information.

Python

To use Amazon Bedrock embedding API, you must have boto3 Python package installed and create an instance of boto3.Session with your AWS credentials. To use:

import boto3
import chromadb.utils.embedding_functions as embedding_functions

session = boto3.Session(
    aws_access_key_id=aws_access_key_id,
    aws_secret_access_key=aws_secret_access_key,
    region_name=region_name,
)
bedrock = embedding_functions.AmazonBedrockEmbeddingFunction(session=session)
bedrock(["document1","document2"])

By default, the embedding function uses the amazon.titan-embed-text-v1 for the model, but you can specify a different model name with model_name parameter. For example:

bedrock = embedding_functions.AmazonBedrockEmbeddingFunction(
    session=session, model_name="cohere.embed-multilingual-v3")
bedrock(["こんにちは", "你们好"])

JavaScript

To use Amazon Bedrock embedding API, you must have @aws-sdk/client-bedrock-runtime installed. To use:

import  { AmazonBedrockEmbeddingFunction } from 'chromadb';
import { fromSSO } = from '@aws-sdk/credential-providers';

const c = await fromSSO({profile: "my-profile"})
const ef = new AmazonBedrockEmbeddingFunction({config: {credentials: c, region: "us-east-1"}})

// use directly
const embeddings = await ef.generate(["foo"])

// pass documents to query for .add and .query
const collection = await client.createCollection({name: "name", embeddingFunction: ef})
const collection = await client.getCollection({name: "name", embeddingFunction: ef})