Generative AI

Weaviate's integration with AWS's SageMaker and Bedrock APIs allows you to access their models' capabilities directly from Weaviate.

Configure a Weaviate collection to use a generative AI model with AWS. Weaviate will perform retrieval augmented generation (RAG) using the specified model and your AWS API credentials.

More specifically, Weaviate will perform a search, retrieve the most relevant objects, and then pass them to the AWS generative model to generate outputs.

Your Weaviate instance must be configured with the AWS generative AI integration (generative-aws) module.

You must provide access key based AWS credentials to Weaviate for these integrations. Go to AWS to sign up and obtain an AWS access key ID and a corresponding AWS secret access key.

Provide the API credentials to Weaviate using one of the following methods:

• Set the AWS_ACCESS_KEY and AWS_SECRET_KEY environment variables that are available to Weaviate.
• Provide the API credentials at runtime, as shown in the examples below.

# Recommended: save sensitive data as environment variables
aws_access_key = os.getenv("AWS_ACCESS_KEY")
aws_secret_key = os.getenv("AWS_SECRET_KEY")

const aws_access_key = process.env.AWS_ACCESS_KEY || '';  // Replace with your AWS access key
const aws_secret_key = process.env.AWS_SECRET_KEY || '';  // Replace with your AWS secret key

To use a model via Bedrock, it must be available, and AWS must grant you access to it.

Refer to the AWS documentation for the list of available models, and to this document to find out how request access to a model.

To use a model via SageMaker, you must have access to the model's endpoint.

Configure a Weaviate index as follows to use an AWS generative model:

For Bedrock, you must provide the model name in the generative AI configuration.

from weaviate.classes.config import Configureclient.collections.create(    "DemoCollection",    generative_config=Configure.Generative.aws(        region="us-east-1",        service="bedrock",        model="cohere.command-r-plus-v1:0"    ))

await client.collections.create({  name: 'DemoCollection',  generative: weaviate.configure.generative.aws({    region: 'us-east-1',    service: 'bedrock',    model: 'cohere.command-r-plus-v1:0',  }),})

For SageMaker, you must provide the endpoint address in the generative AI configuration.

from weaviate.classes.config import Configureclient.collections.create(    "DemoCollection",    generative_config=Configure.Generative.aws(        region="us-east-1",        service="sagemaker",        endpoint="<custom_sagemaker_url>"    ))

await client.collections.create({  name: 'DemoCollection',  generative: weaviate.configure.generative.aws({    region: 'us-east-1',    service: 'sagemaker',    endpoint: '<custom_sagemaker_url>'  }),})

You can specify which model Weaviate uses.

For further details on model parameters, see the relevant AWS documentation.

Aside from setting the default model provider when creating the collection, you can also override it at query time.

from weaviate.classes.config import Configurefrom weaviate.classes.generate import GenerativeConfigcollection = client.collections.use("DemoCollection")response = collection.generate.near_text(    query="A holiday film",    limit=2,    grouped_task="Write a tweet promoting these two movies",    generative_provider=GenerativeConfig.aws(        region="us-east-1",        service="bedrock", # You can also use sagemaker        model="cohere.command-r-plus-v1:0"    ),    # Additional parameters not shown)

import { generativeParameters } from 'weaviate-client';

After configuring the generative AI integration, perform RAG operations, either with the single prompt or grouped task method.

To generate text for each object in the search results, use the single prompt method.

The example below generates outputs for each of the n search results, where n is specified by the limit parameter.

When creating a single prompt query, use braces {} to interpolate the object properties you want Weaviate to pass on to the language model. For example, to pass on the object's title property, include {title} in the query.

collection = client.collections.use("DemoCollection")response = collection.generate.near_text(    query="A holiday film",  # The model provider integration will automatically vectorize the query    single_prompt="Translate this into French: {title}",    limit=2)for obj in response.objects:    print(obj.properties["title"])    print(f"Generated output: {obj.generated}")  # Note that the generated output is per object

let response;
const myCollection = client.collections.use("DemoCollection");

To generate one text for the entire set of search results, use the grouped task method.

In other words, when you have n search results, the generative model generates one output for the entire group.

collection = client.collections.use("DemoCollection")response = collection.generate.near_text(    query="A holiday film",  # The model provider integration will automatically vectorize the query    grouped_task="Write a fun tweet to promote readers to check out these films.",    limit=2)print(f"Generated output: {response.generative.text}")  # Note that the generated output is per queryfor obj in response.objects:    print(obj.properties["title"])

let response;
const myCollection = client.collections.use("DemoCollection");

You can also supply images as a part of the input when performing retrieval augmented generation in both single prompts and grouped tasks.

import base64import requestsfrom weaviate.classes.generate import GenerativeConfig, GenerativeParameterssrc_img_path = "https://upload.wikimedia.org/wikipedia/commons/thumb/b/b0/Winter_forest_silver.jpg/960px-Winter_forest_silver.jpg"base64_image = base64.b64encode(requests.get(src_img_path).content).decode('utf-8')prompt = GenerativeParameters.grouped_task(    prompt="Which movie is closest to the image in terms of atmosphere",    images=[base64_image],      # A list of base64 encoded strings of the image bytes    # image_properties=["img"], # Properties containing images in Weaviate)jeopardy = client.collections.use("DemoCollection")response = jeopardy.generate.near_text(    query="Movies",    limit=5,    grouped_task=prompt,    generative_provider=GenerativeConfig.aws(        region="us-east-1",        service="bedrock", # You can also use sagemaker        model="cohere.command-r-plus-v1:0"    ),)# Print the source property and the generated responsefor o in response.objects:    print(f"Title property: {o.properties['title']}")print(f"Grouped task result: {response.generative.text}")

import { generativeParameters } from 'weaviate-client';

Weaviate passes the model value through to Amazon Bedrock, so any Bedrock text generation model that your AWS account and region has access to can be used. Weaviate recognizes the model families offered by AI21 Labs, Amazon (Titan and Nova), Anthropic, Cohere, Meta, and Mistral AI, including their cross-region inference profile IDs.

For the current model IDs, see the Amazon Bedrock supported foundation models documentation. Refer to this document to find out how to request access to a model.

Any custom SageMaker URL can be used as an endpoint.

• AWS embedding models + Weaviate.

Once the integrations are configured at the collection, the data management and search operations in Weaviate work identically to any other collection. See the following model-agnostic examples:

• The How-to: Manage collections and How-to: Manage objects guides show how to perform data operations (i.e. create, read, update, delete collections and objects within them).
• The How-to: Query & Search guides show how to perform search operations (i.e. vector, keyword, hybrid) as well as retrieval augmented generation.

• AWS Bedrock documentation
• AWS SageMaker documentation

Have a question or feedback? Here's how to reach us.