Skip to main content

Bedrock Embedding

Supported Embedding Models​

ProviderLiteLLM RouteAWS DocumentationCost Tracking
Amazon Titanbedrock/amazon.titan-*Amazon Titan Embeddings✅
Amazon Novabedrock/amazon.nova-*Amazon Nova Embeddings✅
Coherebedrock/cohere.*Cohere Embeddings✅
TwelveLabsbedrock/twelvelabs.*, bedrock/us.twelvelabs.*, bedrock/eu.twelvelabs.*TwelveLabs, Marengo Embed 3.0✅

Async Invoke Support​

LiteLLM supports AWS Bedrock's async-invoke feature for embedding models that require asynchronous processing, particularly useful for large media files (video, audio) or when you need to process embeddings in the background.

Supported Models​

ProviderAsync Invoke RouteUse Case
Amazon Novabedrock/async_invoke/amazon.nova-2-multimodal-embeddings-v1:0Multimodal embeddings with segmentation for long text, video, and audio
TwelveLabs Marengo Embed 2.7bedrock/async_invoke/twelvelabs.marengo-embed-2-7-v1:0Video, audio, image, and text embeddings
TwelveLabs Marengo Embed 3.0bedrock/async_invoke/twelvelabs.marengo-embed-3-0-v1:0Video and audio embeddings (text and image run synchronously, see Marengo Embed 3.0)

Required Parameters​

Bedrock's async invoke only accepts base model ids, so use twelvelabs.marengo-embed-2-7-v1:0 or twelvelabs.marengo-embed-3-0-v1:0 here rather than a us. or eu. inference profile, which Bedrock rejects with The provided model doesn't support async inference

When using async-invoke, you must provide:

ParameterDescriptionRequired
output_s3_uriS3 URI where the embedding results will be stored✅ Yes
input_typeType of input: "text", "image", "video", or "audio" (Marengo Embed 3.0 also takes "text_image" and "multi_input")✅ Yes
aws_region_nameAWS region for the request✅ Yes

Usage​

Basic Async Invoke​

from litellm import embedding

# Text embedding with async-invoke
response = embedding(
    model="bedrock/async_invoke/twelvelabs.marengo-embed-2-7-v1:0",
    input=["Hello world from LiteLLM async invoke!"],
    aws_region_name="us-east-1",
    input_type="text",
    output_s3_uri="s3://your-bucket/async-invoke-output/"
)

print(f"Job submitted! Invocation ARN: {response._hidden_params._invocation_arn}")

Video/Audio Embedding​

# Video embedding (requires async-invoke)
response = embedding(
    model="bedrock/async_invoke/twelvelabs.marengo-embed-2-7-v1:0",
    input=["s3://your-bucket/video.mp4"],  # S3 URL for video
    aws_region_name="us-east-1",
    input_type="video",
    output_s3_uri="s3://your-bucket/async-invoke-output/"
)

print(f"Video embedding job submitted! ARN: {response._hidden_params._invocation_arn}")

Image Embedding with Base64​

import base64

# Load and encode image
with open("image.jpg", "rb") as img_file:
    img_data = base64.b64encode(img_file.read()).decode('utf-8')
    img_base64 = f"data:image/jpeg;base64,{img_data}"

response = embedding(
    model="bedrock/async_invoke/twelvelabs.marengo-embed-2-7-v1:0",
    input=[img_base64],
    aws_region_name="us-east-1",
    input_type="image",
    output_s3_uri="s3://your-bucket/async-invoke-output/"
)

Retrieving Job Information​

Getting Job ID and Invocation ARN​

The async-invoke response includes the invocation ARN in the hidden parameters:

response = embedding(
    model="bedrock/async_invoke/twelvelabs.marengo-embed-2-7-v1:0",
    input=["Hello world"],
    aws_region_name="us-east-1",
    input_type="text",
    output_s3_uri="s3://your-bucket/async-invoke-output/"
)

# Access invocation ARN
invocation_arn = response._hidden_params._invocation_arn
print(f"Invocation ARN: {invocation_arn}")

# Extract job ID from ARN (last part after the last slash)
job_id = invocation_arn.split("/")[-1]
print(f"Job ID: {job_id}")

Checking Job Status​

Use LiteLLM's retrieve_batch function to check if your job is still processing:

from litellm import retrieve_batch

def check_async_job_status(invocation_arn, aws_region_name="us-east-1"):
    """Check the status of an async invoke job using LiteLLM batch API"""
    try:
        response = retrieve_batch(
            batch_id=invocation_arn,  # Pass the invocation ARN here
            custom_llm_provider="bedrock",
            aws_region_name=aws_region_name
        )
        return response
    except Exception as e:
        print(f"Error checking job status: {e}")
        return None

# Check status
status = check_async_job_status(invocation_arn, "us-east-1")
if status:
    print(f"Job Status: {status.status}")  # "in_progress", "completed", or "failed"
    print(f"Output Location: {status.metadata['output_file_id']}")  # S3 URI where results are stored

Polling Until Complete​

Here's a complete example of polling for job completion:

def wait_for_async_job(invocation_arn, aws_region_name="us-east-1", max_wait=3600):
    """Poll job status until completion"""
    start_time = time.time()
    
    while True:
        status = retrieve_batch(
            batch_id=invocation_arn,
            custom_llm_provider="bedrock",
            aws_region_name=aws_region_name,
        )
        
        if status.status == "completed":
            print("✅ Job completed!")
            return status
        elif status.status == "failed":
            error_msg = status.metadata.get('failure_message', 'Unknown error')
            raise Exception(f"❌ Job failed: {error_msg}")
        else:
            elapsed = time.time() - start_time
            if elapsed > max_wait:
                raise TimeoutError(f"Job timed out after {max_wait} seconds")
            
            print(f"⏳ Job still processing... (elapsed: {elapsed:.0f}s)")
            time.sleep(10)  # Wait 10 seconds before checking again

# Wait for completion
completed_status = wait_for_async_job(invocation_arn)
output_s3_uri = completed_status.metadata['output_file_id']
print(f"Results available at: {output_s3_uri}")

Note: The actual embedding results are stored in S3. When the job is completed, download the results from the S3 location specified in status.metadata['output_file_id']. The results will be in JSON/JSONL format containing the embedding vectors.

TwelveLabs Marengo Embed 3.0​

Marengo Embed 3.0 (twelvelabs.marengo-embed-3-0-v1:0, with the us. and eu. inference profiles for us-east-1 and eu-west-1) returns 512-dimensional vectors and takes text of up to 500 tokens. Text, image, text plus image, and multi-image inputs run synchronously through bedrock/<model id>; video and audio go through bedrock/async_invoke/twelvelabs.marengo-embed-3-0-v1:0 with an output_s3_uri, using the base model id because Bedrock's async invoke rejects inference profiles. 3.0 sends Bedrock a different request body than 2.7, and LiteLLM builds it from the same input_type parameter, so moving from 2.7 (scheduled for deprecation on 2026-11-30) is a model id change

Input types​

input_typeinputExtra parameters
text (default)The text to embed
imageA base64 image or an s3:// uri
text_imageThe textmedia_source: a base64 image or an s3:// uri
multi_inputThe text, referencing each image as <@name>media_sources: a mapping of name to base64 image or s3:// uri
video, audioAn s3:// uri or base64 media, async invoke onlystartSec, endSec, segmentation, embeddingOption, embeddingType, embeddingScope, as documented by AWS

inferenceId and bucketOwner (for an s3:// uri owned by another account) are passed through on every input type. A text_image request without media_source, or a multi_input request without media_sources, is rejected with a 400 naming the missing parameter. Marengo 2.7's textTruncate, lengthSec, useFixedLengthSec, and minClipSec, and the video and audio options on any other input type, are rejected with a 400 on 3.0 unless drop_params is set, which drops them instead

Proxy config​

model_list:
  - model_name: marengo-3
    litellm_params:
      model: bedrock/us.twelvelabs.marengo-embed-3-0-v1:0
      aws_region_name: us-east-1
  - model_name: marengo-3-async
    litellm_params:
      model: bedrock/async_invoke/twelvelabs.marengo-embed-3-0-v1:0
      aws_region_name: us-east-1

Text​

curl http://0.0.0.0:4000/v1/embeddings \
  -H "Authorization: Bearer sk-1234" -H "Content-Type: application/json" \
  -d '{"model": "marengo-3", "input": "a dog running on the beach", "input_type": "text"}'

Image​

curl http://0.0.0.0:4000/v1/embeddings \
  -H "Authorization: Bearer sk-1234" -H "Content-Type: application/json" \
  -d '{"model": "marengo-3", "input": "<base64 image>", "input_type": "image"}'

Text plus image​

curl http://0.0.0.0:4000/v1/embeddings \
  -H "Authorization: Bearer sk-1234" -H "Content-Type: application/json" \
  -d '{"model": "marengo-3", "input": "a duck", "input_type": "text_image", "media_source": "s3://your-bucket/duck.png"}'

Multiple images​

curl http://0.0.0.0:4000/v1/embeddings \
  -H "Authorization: Bearer sk-1234" -H "Content-Type: application/json" \
  -d '{"model": "marengo-3", "input": "a photo of <@bird> on water", "input_type": "multi_input", "media_sources": {"bird": "<base64 image>"}}'

Video through async invoke​

from litellm import embedding

response = embedding(
    model="bedrock/async_invoke/twelvelabs.marengo-embed-3-0-v1:0",
    input=["s3://your-bucket/video.mp4"],
    aws_region_name="us-east-1",
    input_type="video",
    embeddingOption=["visual", "audio"],
    segmentation={"method": "fixed", "fixed": {"durationSec": 10}},
    output_s3_uri="s3://your-bucket/async-invoke-output/",
)
print(response._hidden_params._invocation_arn)

Amazon Nova Multimodal Embeddings​

Amazon Nova supports multimodal embeddings for text, images, video, and audio. It offers flexible embedding dimensions and purposes optimized for different use cases.

Supported Features​

  • Modalities: Text, Image, Video, Audio
  • Dimensions: 256, 384, 1024, 3072 (default: 3072)
  • Embedding Purposes:
    • GENERIC_INDEX (default)
    • GENERIC_RETRIEVAL
    • TEXT_RETRIEVAL
    • IMAGE_RETRIEVAL
    • VIDEO_RETRIEVAL
    • AUDIO_RETRIEVAL
    • CLASSIFICATION
    • CLUSTERING

Text Embedding​

from litellm import embedding

response = embedding(
    model="bedrock/amazon.nova-2-multimodal-embeddings-v1:0",
    input=["Hello, world!"],
    aws_region_name="us-east-1",
    dimensions=1024,  # Optional: 256, 384, 1024, or 3072
)

print(response.data[0].embedding)

Image Embedding with Base64​

Amazon Nova accepts images in base64 format using the standard data URL format:

import base64
from litellm import embedding

# Method 1: Load image from file
with open("image.jpg", "rb") as image_file:
    image_data = base64.b64encode(image_file.read()).decode('utf-8')
    # Create data URL with proper format
    image_base64 = f"data:image/jpeg;base64,{image_data}"

response = embedding(
    model="bedrock/amazon.nova-2-multimodal-embeddings-v1:0",
    input=[image_base64],
    aws_region_name="us-east-1",
    dimensions=1024,
)

print(f"Image embedding: {response.data[0].embedding[:10]}...")  # First 10 dimensions

Supported Image Formats​

Nova supports the following image formats:

  • JPEG: data:image/jpeg;base64,...
  • PNG: data:image/png;base64,...
  • GIF: data:image/gif;base64,...
  • WebP: data:image/webp;base64,...

Complete Example with Error Handling​

import base64
from litellm import embedding

def get_image_embedding(image_path, dimensions=1024):
    """
    Get embedding for an image file.
    
    Args:
        image_path: Path to the image file
        dimensions: Embedding dimension (256, 384, 1024, or 3072)
    
    Returns:
        List of embedding values
    """
    try:
        # Determine image format from file extension
        if image_path.lower().endswith('.png'):
            mime_type = "image/png"
        elif image_path.lower().endswith(('.jpg', '.jpeg')):
            mime_type = "image/jpeg"
        elif image_path.lower().endswith('.gif'):
            mime_type = "image/gif"
        elif image_path.lower().endswith('.webp'):
            mime_type = "image/webp"
        else:
            raise ValueError(f"Unsupported image format: {image_path}")
        
        # Read and encode image
        with open(image_path, "rb") as image_file:
            image_data = base64.b64encode(image_file.read()).decode('utf-8')
            image_base64 = f"data:{mime_type};base64,{image_data}"
        
        # Get embedding
        response = embedding(
            model="bedrock/amazon.nova-2-multimodal-embeddings-v1:0",
            input=[image_base64],
            aws_region_name="us-east-1",
            dimensions=dimensions,
        )
        
        return response.data[0].embedding
        
    except Exception as e:
        print(f"Error getting image embedding: {e}")
        raise

# Example usage
image_embedding = get_image_embedding("photo.jpg", dimensions=1024)
print(f"Got embedding with {len(image_embedding)} dimensions")

Error Handling​

Common Errors​

ErrorCauseSolution
ValueError: output_s3_uri cannot be emptyMissing S3 output URIProvide a valid S3 URI
ValueError: Input type 'video' requires async_invoke routeUsing video/audio without async-invokeUse bedrock/async_invoke/ model prefix
ValueError: input_type is requiredMissing input type parameterSpecify input_type parameter

Example Error Handling​

try:
    response = embedding(
        model="bedrock/async_invoke/twelvelabs.marengo-embed-2-7-v1:0",
        input=["Hello world"],
        aws_region_name="us-east-1",
        input_type="text",
        output_s3_uri="s3://your-bucket/output/"  # Required for async-invoke
    )
    print("Job submitted successfully!")
    
except ValueError as e:
    if "output_s3_uri cannot be empty" in str(e):
        print("Error: Please provide a valid S3 output URI")
    elif "requires async_invoke route" in str(e):
        print("Error: Use async_invoke model for video/audio inputs")
    else:
        print(f"Error: {e}")
except Exception as e:
    print(f"Unexpected error: {e}")

Best Practices​

  1. Use async-invoke for large files: Video and audio files are better processed asynchronously
  2. Use LiteLLM batch API: Use retrieve_batch() instead of direct Bedrock API calls for status checking
  3. Monitor job status: Check job status periodically using the batch API to know when results are ready
  4. Handle errors gracefully: Implement proper error handling for network issues and job failures
  5. Set appropriate timeouts: Consider the processing time for large files
  6. Use S3 for large inputs: For video/audio, use S3 URLs instead of base64 encoding

Limitations​

  • Async-invoke is supported for TwelveLabs Marengo and Amazon Nova models
  • Results are stored in S3 and must be retrieved separately using the output file ID
  • Job status checking requires using LiteLLM's retrieve_batch() function
  • No built-in polling mechanism in LiteLLM (must implement your own status checking loop)

API keys​

This can be set as env variables or passed as params to litellm.embedding()

import os
os.environ["AWS_ACCESS_KEY_ID"] = ""        # Access key
os.environ["AWS_SECRET_ACCESS_KEY"] = ""    # Secret access key
os.environ["AWS_REGION_NAME"] = ""           # us-east-1, us-east-2, us-west-1, us-west-2

Usage​

LiteLLM Python SDK​

from litellm import embedding
response = embedding(
    model="bedrock/amazon.titan-embed-text-v1",
    input=["good morning from litellm"],
)
print(response)

LiteLLM Proxy Server​

1. Setup config.yaml​

model_list:
  - model_name: titan-embed-v1
    litellm_params:
      model: bedrock/amazon.titan-embed-text-v1
      aws_access_key_id: os.environ/AWS_ACCESS_KEY_ID
      aws_secret_access_key: os.environ/AWS_SECRET_ACCESS_KEY
      aws_region_name: us-east-1
  - model_name: titan-embed-v2
    litellm_params:
      model: bedrock/amazon.titan-embed-text-v2:0
      aws_access_key_id: os.environ/AWS_ACCESS_KEY_ID
      aws_secret_access_key: os.environ/AWS_SECRET_ACCESS_KEY
      aws_region_name: us-east-1

2. Start Proxy​

litellm --config /path/to/config.yaml

3. Use with OpenAI Python SDK​

import openai
client = openai.OpenAI(
    api_key="anything",
    base_url="http://0.0.0.0:4000"
)

response = client.embeddings.create(
    input=["good morning from litellm"],
    model="titan-embed-v1"
)
print(response)

4. Use with LiteLLM Python SDK​

import litellm
response = litellm.embedding(
    model="titan-embed-v1", # model alias from config.yaml
    input=["good morning from litellm"],
    api_base="http://0.0.0.0:4000",
    api_key="anything"
)
print(response)

Supported AWS Bedrock Embedding Models​

Model NameUsageSupported Additional OpenAI params
Amazon Nova Multimodal Embeddingsembedding(model="bedrock/amazon.nova-2-multimodal-embeddings-v1:0", input=input)Supports multimodal input (text, image, video, audio), multiple purposes, dimensions (256, 384, 1024, 3072)
Titan Embeddings V2embedding(model="bedrock/amazon.titan-embed-text-v2:0", input=input)here
Titan Embeddings - V1embedding(model="bedrock/amazon.titan-embed-text-v1", input=input)here
Titan Multimodal Embeddingsembedding(model="bedrock/amazon.titan-embed-image-v1", input=input)here
TwelveLabs Marengo Embed 2.7embedding(model="bedrock/us.twelvelabs.marengo-embed-2-7-v1:0", input=input)Supports multimodal input (text, video, audio, image)
TwelveLabs Marengo Embed 3.0embedding(model="bedrock/us.twelvelabs.marengo-embed-3-0-v1:0", input=input)Text, image, text_image, and multi_input synchronously, video and audio through async invoke, 512 dimensions
Cohere Embeddings - Englishembedding(model="bedrock/cohere.embed-english-v3", input=input)here
Cohere Embeddings - Multilingualembedding(model="bedrock/cohere.embed-multilingual-v3", input=input)here
Cohere Embed v4embedding(model="bedrock/cohere.embed-v4:0", input=input)Supports text and image input, configurable dimensions (256, 512, 1024, 1536), 128k context length

Advanced - Drop Unsupported Params​

Advanced - Pass model/provider-specific Params​

🚅
LiteLLM Enterprise
SSO/SAML, audit logs, spend tracking, multi-team management, and guardrails — built for production.
Learn more →