Skip to main content

AWS Bedrock

Anthropic, Amazon Titan, A121 LLMs are Supported on Bedrock

LiteLLM requires boto3 to be installed on your system for Bedrock requests

pip install boto3>=1.28.57

Required Environment Variables

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

Open In Colab
import os
from litellm import completion

os.environ["AWS_ACCESS_KEY_ID"] = ""
os.environ["AWS_SECRET_ACCESS_KEY"] = ""
os.environ["AWS_REGION_NAME"] = ""

response = completion(
model="bedrock/anthropic.claude-3-sonnet-20240229-v1:0",
messages=[{ "content": "Hello, how are you?","role": "user"}]
)

OpenAI Proxy Usage

Here's how to call Anthropic with the LiteLLM Proxy Server

1. Save key in your environment

export AWS_ACCESS_KEY_ID=""
export AWS_SECRET_ACCESS_KEY=""
export AWS_REGION_NAME=""

2. Start the proxy

$ litellm --model anthropic.claude-3-sonnet-20240229-v1:0

# Server running on http://0.0.0.0:4000

3. Test it

curl --location 'http://0.0.0.0:4000/chat/completions' \
--header 'Content-Type: application/json' \
--data ' {
"model": "bedrock-claude-v1",
"messages": [
{
"role": "user",
"content": "what llm are you"
}
]
}
'

Usage - Function Calling

info

Claude returns it's output as an XML Tree. Here is how we translate it.

You can see the raw response via response._hidden_params["original_response"].

Claude hallucinates, e.g. returning the list param value as <value>\n<item>apple</item>\n<item>banana</item>\n</value> or <value>\n<list>\n<item>apple</item>\n<item>banana</item>\n</list>\n</value>.

from litellm import completion

# set env
os.environ["AWS_ACCESS_KEY_ID"] = ""
os.environ["AWS_SECRET_ACCESS_KEY"] = ""
os.environ["AWS_REGION_NAME"] = ""

tools = [
{
"type": "function",
"function": {
"name": "get_current_weather",
"description": "Get the current weather in a given location",
"parameters": {
"type": "object",
"properties": {
"location": {
"type": "string",
"description": "The city and state, e.g. San Francisco, CA",
},
"unit": {"type": "string", "enum": ["celsius", "fahrenheit"]},
},
"required": ["location"],
},
},
}
]
messages = [{"role": "user", "content": "What's the weather like in Boston today?"}]

response = completion(
model="bedrock/anthropic.claude-3-sonnet-20240229-v1:0",
messages=messages,
tools=tools,
tool_choice="auto",
)
# Add any assertions, here to check response args
print(response)
assert isinstance(response.choices[0].message.tool_calls[0].function.name, str)
assert isinstance(
response.choices[0].message.tool_calls[0].function.arguments, str
)

Usage - Vision

from litellm import completion

# set env
os.environ["AWS_ACCESS_KEY_ID"] = ""
os.environ["AWS_SECRET_ACCESS_KEY"] = ""
os.environ["AWS_REGION_NAME"] = ""


def encode_image(image_path):
import base64

with open(image_path, "rb") as image_file:
return base64.b64encode(image_file.read()).decode("utf-8")


image_path = "../proxy/cached_logo.jpg"
# Getting the base64 string
base64_image = encode_image(image_path)
resp = litellm.completion(
model="bedrock/anthropic.claude-3-sonnet-20240229-v1:0",
messages=[
{
"role": "user",
"content": [
{"type": "text", "text": "Whats in this image?"},
{
"type": "image_url",
"image_url": {
"url": "data:image/jpeg;base64," + base64_image
},
},
],
}
],
)
print(f"\nResponse: {resp}")

Usage - "Assistant Pre-fill"

If you're using Anthropic's Claude with Bedrock, you can "put words in Claude's mouth" by including an assistant role message as the last item in the messages array.

[!IMPORTANT] The returned completion will not include your "pre-fill" text, since it is part of the prompt itself. Make sure to prefix Claude's completion with your pre-fill.

import os
from litellm import completion

os.environ["AWS_ACCESS_KEY_ID"] = ""
os.environ["AWS_SECRET_ACCESS_KEY"] = ""
os.environ["AWS_REGION_NAME"] = ""

messages = [
{"role": "user", "content": "How do you say 'Hello' in German? Return your answer as a JSON object, like this:\n\n{ \"Hello\": \"Hallo\" }"},
{"role": "assistant", "content": "{"},
]
response = completion(model="bedrock/anthropic.claude-v2", messages=messages)

Example prompt sent to Claude


Human: How do you say 'Hello' in German? Return your answer as a JSON object, like this:

{ "Hello": "Hallo" }

Assistant: {

Usage - "System" messages

If you're using Anthropic's Claude 2.1 with Bedrock, system role messages are properly formatted for you.

import os
from litellm import completion

os.environ["AWS_ACCESS_KEY_ID"] = ""
os.environ["AWS_SECRET_ACCESS_KEY"] = ""
os.environ["AWS_REGION_NAME"] = ""

messages = [
{"role": "system", "content": "You are a snarky assistant."},
{"role": "user", "content": "How do I boil water?"},
]
response = completion(model="bedrock/anthropic.claude-v2:1", messages=messages)

Example prompt sent to Claude

You are a snarky assistant.

Human: How do I boil water?

Assistant:

Usage - Streaming

import os
from litellm import completion

os.environ["AWS_ACCESS_KEY_ID"] = ""
os.environ["AWS_SECRET_ACCESS_KEY"] = ""
os.environ["AWS_REGION_NAME"] = ""

response = completion(
model="bedrock/anthropic.claude-instant-v1",
messages=[{ "content": "Hello, how are you?","role": "user"}],
stream=True
)
for chunk in response:
print(chunk)

Example Streaming Output Chunk

{
"choices": [
{
"finish_reason": null,
"index": 0,
"delta": {
"content": "ase can appeal the case to a higher federal court. If a higher federal court rules in a way that conflicts with a ruling from a lower federal court or conflicts with a ruling from a higher state court, the parties involved in the case can appeal the case to the Supreme Court. In order to appeal a case to the Sup"
}
}
],
"created": null,
"model": "anthropic.claude-instant-v1",
"usage": {
"prompt_tokens": null,
"completion_tokens": null,
"total_tokens": null
}
}

Boto3 - Authentication

Passing credentials as parameters - Completion()

Pass AWS credentials as parameters to litellm.completion

import os
from litellm import completion

response = completion(
model="bedrock/anthropic.claude-instant-v1",
messages=[{ "content": "Hello, how are you?","role": "user"}],
aws_access_key_id="",
aws_secret_access_key="",
aws_region_name="",
)

Passing an external BedrockRuntime.Client as a parameter - Completion()

Pass an external BedrockRuntime.Client object as a parameter to litellm.completion. Useful when using an AWS credentials profile, SSO session, assumed role session, or if environment variables are not available for auth.

Create a client from session credentials:

import boto3
from litellm import completion

bedrock = boto3.client(
service_name="bedrock-runtime",
region_name="us-east-1",
aws_access_key_id="",
aws_secret_access_key="",
aws_session_token="",
)

response = completion(
model="bedrock/anthropic.claude-instant-v1",
messages=[{ "content": "Hello, how are you?","role": "user"}],
aws_bedrock_client=bedrock,
)

Create a client from AWS profile in ~/.aws/config:

import boto3
from litellm import completion

dev_session = boto3.Session(profile_name="dev-profile")
bedrock = dev_session.client(
service_name="bedrock-runtime",
region_name="us-east-1",
)

response = completion(
model="bedrock/anthropic.claude-instant-v1",
messages=[{ "content": "Hello, how are you?","role": "user"}],
aws_bedrock_client=bedrock,
)

SSO Login (AWS Profile)

  • Set AWS_PROFILE environment variable
  • Make bedrock completion call
import os
from litellm import completion

response = completion(
model="bedrock/anthropic.claude-instant-v1",
messages=[{ "content": "Hello, how are you?","role": "user"}]
)

or pass aws_profile_name:

import os
from litellm import completion

response = completion(
model="bedrock/anthropic.claude-instant-v1",
messages=[{ "content": "Hello, how are you?","role": "user"}],
aws_profile_name="dev-profile",
)

STS based Auth

  • Set aws_role_name and aws_session_name in completion() / embedding() function

Make the bedrock completion call

from litellm import completion

response = completion(
model="bedrock/anthropic.claude-instant-v1",
messages=messages,
max_tokens=10,
temperature=0.1,
aws_role_name=aws_role_name,
aws_session_name="my-test-session",
)

If you also need to dynamically set the aws user accessing the role, add the additional args in the completion()/embedding() function

from litellm import completion

response = completion(
model="bedrock/anthropic.claude-instant-v1",
messages=messages,
max_tokens=10,
temperature=0.1,
aws_region_name=aws_region_name,
aws_access_key_id=aws_access_key_id,
aws_secret_access_key=aws_secret_access_key,
aws_role_name=aws_role_name,
aws_session_name="my-test-session",
)

Provisioned throughput models

To use provisioned throughput Bedrock models pass

  • model=bedrock/<base-model>, example model=bedrock/anthropic.claude-v2. Set model to any of the Supported AWS models
  • model_id=provisioned-model-arn

Completion

import litellm
response = litellm.completion(
model="bedrock/anthropic.claude-instant-v1",
model_id="provisioned-model-arn",
messages=[{"content": "Hello, how are you?", "role": "user"}]
)

Embedding

import litellm
response = litellm.embedding(
model="bedrock/amazon.titan-embed-text-v1",
model_id="provisioned-model-arn",
input=["hi"],
)

Supported AWS Bedrock Models

Here's an example of using a bedrock model with LiteLLM

Model NameCommand
Anthropic Claude-V3 sonnetcompletion(model='bedrock/anthropic.claude-3-sonnet-20240229-v1:0', messages=messages)
Anthropic Claude-V3 Haikucompletion(model='bedrock/anthropic.claude-3-haiku-20240307-v1:0', messages=messages)
Anthropic Claude-V2.1completion(model='bedrock/anthropic.claude-v2:1', messages=messages)
Anthropic Claude-V2completion(model='bedrock/anthropic.claude-v2', messages=messages)
Anthropic Claude-Instant V1completion(model='bedrock/anthropic.claude-instant-v1', messages=messages)
Amazon Titan Litecompletion(model='bedrock/amazon.titan-text-lite-v1', messages=messages)
Amazon Titan Expresscompletion(model='bedrock/amazon.titan-text-express-v1', messages=messages)
Cohere Commandcompletion(model='bedrock/cohere.command-text-v14', messages=messages)
AI21 J2-Midcompletion(model='bedrock/ai21.j2-mid-v1', messages=messages)
AI21 J2-Ultracompletion(model='bedrock/ai21.j2-ultra-v1', messages=messages)
Meta Llama 2 Chat 13bcompletion(model='bedrock/meta.llama2-13b-chat-v1', messages=messages)
Meta Llama 2 Chat 70bcompletion(model='bedrock/meta.llama2-70b-chat-v1', messages=messages)
Mistral 7B Instructcompletion(model='bedrock/mistral.mistral-7b-instruct-v0:2', messages=messages)
Mixtral 8x7B Instructcompletion(model='bedrock/mistral.mixtral-8x7b-instruct-v0:1', messages=messages)

Bedrock Embedding

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

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

Supported AWS Bedrock Embedding Models

Model NameFunction Call
Titan Embeddings - G1embedding(model="bedrock/amazon.titan-embed-text-v1", input=input)
Cohere Embeddings - Englishembedding(model="bedrock/cohere.embed-english-v3", input=input)
Cohere Embeddings - Multilingualembedding(model="bedrock/cohere.embed-multilingual-v3", input=input)

Image Generation

Use this for stable diffusion on bedrock

Usage

import os
from litellm import image_generation

os.environ["AWS_ACCESS_KEY_ID"] = ""
os.environ["AWS_SECRET_ACCESS_KEY"] = ""
os.environ["AWS_REGION_NAME"] = ""

response = image_generation(
prompt="A cute baby sea otter",
model="bedrock/stability.stable-diffusion-xl-v0",
)
print(f"response: {response}")

Set optional params

import os
from litellm import image_generation

os.environ["AWS_ACCESS_KEY_ID"] = ""
os.environ["AWS_SECRET_ACCESS_KEY"] = ""
os.environ["AWS_REGION_NAME"] = ""

response = image_generation(
prompt="A cute baby sea otter",
model="bedrock/stability.stable-diffusion-xl-v0",
### OPENAI-COMPATIBLE ###
size="128x512", # width=128, height=512
### PROVIDER-SPECIFIC ### see `AmazonStabilityConfig` in bedrock.py for all params
seed=30
)
print(f"response: {response}")

Supported AWS Bedrock Image Generation Models

Model NameFunction Call
Stable Diffusion - v0embedding(model="bedrock/stability.stable-diffusion-xl-v0", prompt=prompt)
Stable Diffusion - v0embedding(model="bedrock/stability.stable-diffusion-xl-v1", prompt=prompt)