Skip to content

Latest commit

 

History

History
828 lines (651 loc) · 23.7 KB

vertex.md

File metadata and controls

828 lines (651 loc) · 23.7 KB
Raw

import Image from '@theme/IdealImage'; import Tabs from '@theme/Tabs'; import TabItem from '@theme/TabItem';

VertexAI [Anthropic, Gemini, Model Garden]

Open In Colab

🆕 vertex_ai_beta/ route

New vertex_ai_beta/ route. Adds support for system messages, tool_choice params, etc. by moving to httpx client (instead of vertex sdk).

from litellm import completion
import json 

## GET CREDENTIALS 
file_path = 'path/to/vertex_ai_service_account.json'

# Load the JSON file
with open(file_path, 'r') as file:
    vertex_credentials = json.load(file)

# Convert to JSON string
vertex_credentials_json = json.dumps(vertex_credentials)

## COMPLETION CALL 
response = completion(
  model="vertex_ai_beta/gemini-pro",
  messages=[{ "content": "Hello, how are you?","role": "user"}],
  vertex_credentials=vertex_credentials_json
)

System Message

from litellm import completion
import json 

## GET CREDENTIALS 
file_path = 'path/to/vertex_ai_service_account.json'

# Load the JSON file
with open(file_path, 'r') as file:
    vertex_credentials = json.load(file)

# Convert to JSON string
vertex_credentials_json = json.dumps(vertex_credentials)


response = completion(
  model="vertex_ai_beta/gemini-pro",
  messages=[{"content": "You are a good bot.","role": "system"}, {"content": "Hello, how are you?","role": "user"}], 
  vertex_credentials=vertex_credentials_json
)

Function Calling

Force Gemini to make tool calls with tool_choice="required".

from litellm import completion
import json 

## GET CREDENTIALS 
file_path = 'path/to/vertex_ai_service_account.json'

# Load the JSON file
with open(file_path, 'r') as file:
    vertex_credentials = json.load(file)

# Convert to JSON string
vertex_credentials_json = json.dumps(vertex_credentials)


messages = [
    {
        "role": "system",
        "content": "Your name is Litellm Bot, you are a helpful assistant",
    },
    # User asks for their name and weather in San Francisco
    {
        "role": "user",
        "content": "Hello, what is your name and can you tell me the weather?",
    },
]

tools = [
    {
        "type": "function",
        "function": {
            "name": "get_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",
                    }
                },
                "required": ["location"],
            },
        },
    }
]

data = {
    "model": "vertex_ai_beta/gemini-1.5-pro-preview-0514"),
    "messages": messages,
    "tools": tools,
    "tool_choice": "required",
    "vertex_credentials": vertex_credentials_json
}

## COMPLETION CALL 
print(completion(**data))

JSON Schema

from litellm import completion 

## GET CREDENTIALS 
file_path = 'path/to/vertex_ai_service_account.json'

# Load the JSON file
with open(file_path, 'r') as file:
    vertex_credentials = json.load(file)

# Convert to JSON string
vertex_credentials_json = json.dumps(vertex_credentials)

messages = [
    {
        "role": "user",
        "content": """
List 5 popular cookie recipes.

Using this JSON schema:

    Recipe = {"recipe_name": str}

Return a `list[Recipe]`
        """
    }
]

completion(model="vertex_ai_beta/gemini-1.5-flash-preview-0514", messages=messages, response_format={ "type": "json_object" })

Pre-requisites

  • pip install google-cloud-aiplatform (pre-installed on proxy docker image)

  • Authentication:

    • run gcloud auth application-default login See Google Cloud Docs
    • Alternatively you can set GOOGLE_APPLICATION_CREDENTIALS

    Here's how: Jump to Code

    - Create a service account on GCP
- Export the credentials as a json
- load the json and json.dump the json as a string
- store the json string in your environment as `GOOGLE_APPLICATION_CREDENTIALS`

Sample Usage

import litellm
litellm.vertex_project = "hardy-device-38811" # Your Project ID
litellm.vertex_location = "us-central1"  # proj location

response = litellm.completion(model="gemini-pro", messages=[{"role": "user", "content": "write code for saying hi from LiteLLM"}])

Usage with LiteLLM Proxy Server

Here's how to use Vertex AI with the LiteLLM Proxy Server

  1. Modify the config.yaml

Use this when you need to set a different location for each vertex model

model_list:
  - model_name: gemini-vision
    litellm_params:
      model: vertex_ai/gemini-1.0-pro-vision-001
      vertex_project: "project-id"
      vertex_location: "us-central1"
  - model_name: gemini-vision
    litellm_params:
      model: vertex_ai/gemini-1.0-pro-vision-001
      vertex_project: "project-id2"
      vertex_location: "us-east"

Use this when you have one vertex location for all models

litellm_settings: 
  vertex_project: "hardy-device-38811" # Your Project ID
  vertex_location: "us-central1" # proj location

model_list: 
  -model_name: team1-gemini-pro
  litellm_params: 
    model: gemini-pro
  1. Start the proxy
$ litellm --config /path/to/config.yaml
  1. Send Request to LiteLLM Proxy Server
import openai
client = openai.OpenAI(
    api_key="sk-1234",             # pass litellm proxy key, if you're using virtual keys
    base_url="http://0.0.0.0:4000" # litellm-proxy-base url
)

response = client.chat.completions.create(
    model="team1-gemini-pro",
    messages = [
        {
            "role": "user",
            "content": "what llm are you"
        }
    ],
)

print(response)
curl --location 'http://0.0.0.0:4000/chat/completions' \
    --header 'Authorization: Bearer sk-1234' \
    --header 'Content-Type: application/json' \
    --data '{
    "model": "team1-gemini-pro",
    "messages": [
        {
        "role": "user",
        "content": "what llm are you"
        }
    ],
}'

Specifying Safety Settings

In certain use-cases you may need to make calls to the models and pass safety settigns different from the defaults. To do so, simple pass the safety_settings argument to completion or acompletion. For example:

response = completion(
    model="vertex_ai/gemini-pro", 
    messages=[{"role": "user", "content": "write code for saying hi from LiteLLM"}]
    safety_settings=[
        {
            "category": "HARM_CATEGORY_HARASSMENT",
            "threshold": "BLOCK_NONE",
        },
        {
            "category": "HARM_CATEGORY_HATE_SPEECH",
            "threshold": "BLOCK_NONE",
        },
        {
            "category": "HARM_CATEGORY_SEXUALLY_EXPLICIT",
            "threshold": "BLOCK_NONE",
        },
        {
            "category": "HARM_CATEGORY_DANGEROUS_CONTENT",
            "threshold": "BLOCK_NONE",
        },
    ]
)

Option 1: Set in config

model_list:
  - model_name: gemini-experimental
    litellm_params:
      model: vertex_ai/gemini-experimental
      vertex_project: litellm-epic
      vertex_location: us-central1
      safety_settings:
      - category: HARM_CATEGORY_HARASSMENT
        threshold: BLOCK_NONE
      - category: HARM_CATEGORY_HATE_SPEECH
        threshold: BLOCK_NONE
      - category: HARM_CATEGORY_SEXUALLY_EXPLICIT
        threshold: BLOCK_NONE
      - category: HARM_CATEGORY_DANGEROUS_CONTENT
        threshold: BLOCK_NONE

Option 2: Set on call

response = client.chat.completions.create(
    model="gemini-experimental",
    messages=[
        {
            "role": "user",
            "content": "Can you write exploits?",
        }
    ],
    max_tokens=8192,
    stream=False,
    temperature=0.0,

    extra_body={
        "safety_settings": [
            {
                "category": "HARM_CATEGORY_HARASSMENT",
                "threshold": "BLOCK_NONE",
            },
            {
                "category": "HARM_CATEGORY_HATE_SPEECH",
                "threshold": "BLOCK_NONE",
            },
            {
                "category": "HARM_CATEGORY_SEXUALLY_EXPLICIT",
                "threshold": "BLOCK_NONE",
            },
            {
                "category": "HARM_CATEGORY_DANGEROUS_CONTENT",
                "threshold": "BLOCK_NONE",
            },
        ],
    }
)

Set Vertex Project & Vertex Location

All calls using Vertex AI require the following parameters:

  • Your Project ID
import os, litellm 

# set via env var
os.environ["VERTEXAI_PROJECT"] = "hardy-device-38811" # Your Project ID`

### OR ###

# set directly on module 
litellm.vertex_project = "hardy-device-38811" # Your Project ID`
  • Your Project Location
import os, litellm 

# set via env var
os.environ["VERTEXAI_LOCATION"] = "us-central1 # Your Location

### OR ###

# set directly on module 
litellm.vertex_location = "us-central1 # Your Location

Anthropic

Model Name Function Call
claude-3-opus@20240229 completion('vertex_ai/claude-3-opus@20240229', messages)
claude-3-sonnet@20240229 completion('vertex_ai/claude-3-sonnet@20240229', messages)
claude-3-haiku@20240307 completion('vertex_ai/claude-3-haiku@20240307', messages)

Usage

from litellm import completion
import os

os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = ""

model = "claude-3-sonnet@20240229"

vertex_ai_project = "your-vertex-project" # can also set this as os.environ["VERTEXAI_PROJECT"]
vertex_ai_location = "your-vertex-location" # can also set this as os.environ["VERTEXAI_LOCATION"]

response = completion(
    model="vertex_ai/" + model,
    messages=[{"role": "user", "content": "hi"}],
    temperature=0.7,
    vertex_ai_project=vertex_ai_project,
    vertex_ai_location=vertex_ai_location,
)
print("\nModel Response", response)

1. Add to config

model_list:
    - model_name: anthropic-vertex
      litellm_params:
        model: vertex_ai/claude-3-sonnet@20240229
        vertex_ai_project: "my-test-project"
        vertex_ai_location: "us-east-1"
    - model_name: anthropic-vertex
      litellm_params:
        model: vertex_ai/claude-3-sonnet@20240229
        vertex_ai_project: "my-test-project"
        vertex_ai_location: "us-west-1"

2. Start proxy

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

# RUNNING at http://0.0.0.0:4000

3. Test it!

curl --location 'http://0.0.0.0:4000/chat/completions' \
      --header 'Authorization: Bearer sk-1234' \
      --header 'Content-Type: application/json' \
      --data '{
            "model": "anthropic-vertex", # 👈 the 'model_name' in config
            "messages": [
                {
                "role": "user",
                "content": "what llm are you"
                }
            ],
        }'

Model Garden

Model Name Function Call
llama2 completion('vertex_ai/<endpoint_id>', messages)

Using Model Garden

from litellm import completion
import os

## set ENV variables
os.environ["VERTEXAI_PROJECT"] = "hardy-device-38811"
os.environ["VERTEXAI_LOCATION"] = "us-central1"

response = completion(
  model="vertex_ai/<your-endpoint-id>", 
  messages=[{ "content": "Hello, how are you?","role": "user"}]
)

Gemini Pro

Model Name Function Call
gemini-pro completion('gemini-pro', messages), completion('vertex_ai/gemini-pro', messages)

Gemini Pro Vision

Model Name Function Call
gemini-pro-vision completion('gemini-pro-vision', messages), completion('vertex_ai/gemini-pro-vision', messages)

Gemini 1.5 Pro (and Vision)

Model Name Function Call
gemini-1.5-pro completion('gemini-1.5-pro', messages), completion('vertex_ai/gemini-pro', messages)
gemini-1.5-flash-preview-0514 completion('gemini-1.5-flash-preview-0514', messages), completion('vertex_ai/gemini-pro', messages)
gemini-1.5-pro-preview-0514 completion('gemini-1.5-pro-preview-0514', messages), completion('vertex_ai/gemini-1.5-pro-preview-0514', messages)

Using Gemini Pro Vision

Call gemini-pro-vision in the same input/output format as OpenAI gpt-4-vision

LiteLLM Supports the following image types passed in url

Example Request - image url

import litellm

response = litellm.completion(
  model = "vertex_ai/gemini-pro-vision",
  messages=[
      {
          "role": "user",
          "content": [
                          {
                              "type": "text",
                              "text": "Whats in this image?"
                          },
                          {
                              "type": "image_url",
                              "image_url": {
                              "url": "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg"
                              }
                          }
                      ]
      }
  ],
)
print(response)
import litellm

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 = "cached_logo.jpg"
# Getting the base64 string
base64_image = encode_image(image_path)
response = litellm.completion(
    model="vertex_ai/gemini-pro-vision",
    messages=[
        {
            "role": "user",
            "content": [
                {"type": "text", "text": "Whats in this image?"},
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "data:image/jpeg;base64," + base64_image
                    },
                },
            ],
        }
    ],
)
print(response)

Usage - Function Calling

LiteLLM supports Function Calling for Vertex AI gemini models.

from litellm import completion
import os
# set env
os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = ".."
os.environ["VERTEX_AI_PROJECT"] = ".."
os.environ["VERTEX_AI_LOCATION"] = ".."

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="vertex_ai/gemini-pro-vision",
    messages=messages,
    tools=tools,
)
# 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
)

Chat Models

Model Name Function Call
chat-bison-32k completion('chat-bison-32k', messages)
chat-bison completion('chat-bison', messages)
chat-bison@001 completion('chat-bison@001', messages)

Code Chat Models

Model Name Function Call
codechat-bison completion('codechat-bison', messages)
codechat-bison-32k completion('codechat-bison-32k', messages)
codechat-bison@001 completion('codechat-bison@001', messages)

Text Models

Model Name Function Call
text-bison completion('text-bison', messages)
text-bison@001 completion('text-bison@001', messages)

Code Text Models

Model Name Function Call
code-bison completion('code-bison', messages)
code-bison@001 completion('code-bison@001', messages)
code-gecko@001 completion('code-gecko@001', messages)
code-gecko@latest completion('code-gecko@latest', messages)

Embedding Models

Usage - Embedding

import litellm
from litellm import embedding
litellm.vertex_project = "hardy-device-38811" # Your Project ID
litellm.vertex_location = "us-central1"  # proj location

response = embedding(
    model="vertex_ai/textembedding-gecko",
    input=["good morning from litellm"],
)
print(response)

Supported Embedding Models

All models listed here are supported

Model Name Function Call
text-embedding-004 embedding(model="vertex_ai/text-embedding-004", input)
text-multilingual-embedding-002 embedding(model="vertex_ai/text-multilingual-embedding-002", input)
textembedding-gecko embedding(model="vertex_ai/textembedding-gecko", input)
textembedding-gecko-multilingual embedding(model="vertex_ai/textembedding-gecko-multilingual", input)
textembedding-gecko-multilingual@001 embedding(model="vertex_ai/textembedding-gecko-multilingual@001", input)
textembedding-gecko@001 embedding(model="vertex_ai/textembedding-gecko@001", input)
textembedding-gecko@003 embedding(model="vertex_ai/textembedding-gecko@003", input)
text-embedding-preview-0409 embedding(model="vertex_ai/text-embedding-preview-0409", input)
text-multilingual-embedding-preview-0409 embedding(model="vertex_ai/text-multilingual-embedding-preview-0409", input)

Advanced Use task_type and title (Vertex Specific Params)

👉 task_type and title are vertex specific params

LiteLLM Supported Vertex Specific Params

auto_truncate: Optional[bool] = None
task_type: Optional[Literal["RETRIEVAL_QUERY","RETRIEVAL_DOCUMENT", "SEMANTIC_SIMILARITY", "CLASSIFICATION", "CLUSTERING", "QUESTION_ANSWERING", "FACT_VERIFICATION"]] = None
title: Optional[str] = None # The title of the document to be embedded. (only valid with task_type=RETRIEVAL_DOCUMENT).

Example Usage with LiteLLM

response = litellm.embedding(
    model="vertex_ai/text-embedding-004",
    input=["good morning from litellm", "gm"]
    task_type = "RETRIEVAL_DOCUMENT",
    dimensions=1,
    auto_truncate=True,
)

Image Generation Models

Usage

response = await litellm.aimage_generation(
    prompt="An olympic size swimming pool",
    model="vertex_ai/imagegeneration@006",
    vertex_ai_project="adroit-crow-413218",
    vertex_ai_location="us-central1",
)

Generating multiple images

Use the n parameter to pass how many images you want generated

response = await litellm.aimage_generation(
    prompt="An olympic size swimming pool",
    model="vertex_ai/imagegeneration@006",
    vertex_ai_project="adroit-crow-413218",
    vertex_ai_location="us-central1",
    n=1,
)

Extra

Using GOOGLE_APPLICATION_CREDENTIALS

Here's the code for storing your service account credentials as GOOGLE_APPLICATION_CREDENTIALS environment variable:

def load_vertex_ai_credentials():
  # Define the path to the vertex_key.json file
  print("loading vertex ai credentials")
  filepath = os.path.dirname(os.path.abspath(__file__))
  vertex_key_path = filepath + "/vertex_key.json"

  # Read the existing content of the file or create an empty dictionary
  try:
      with open(vertex_key_path, "r") as file:
          # Read the file content
          print("Read vertexai file path")
          content = file.read()

          # If the file is empty or not valid JSON, create an empty dictionary
          if not content or not content.strip():
              service_account_key_data = {}
          else:
              # Attempt to load the existing JSON content
              file.seek(0)
              service_account_key_data = json.load(file)
  except FileNotFoundError:
      # If the file doesn't exist, create an empty dictionary
      service_account_key_data = {}

  # Create a temporary file
  with tempfile.NamedTemporaryFile(mode="w+", delete=False) as temp_file:
      # Write the updated content to the temporary file
      json.dump(service_account_key_data, temp_file, indent=2)

  # Export the temporary file as GOOGLE_APPLICATION_CREDENTIALS
  os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = os.path.abspath(temp_file.name)

Using GCP Service Account

:::info

Trying to deploy LiteLLM on Google Cloud Run? Tutorial here

:::

  1. Figure out the Service Account bound to the Google Cloud Run service

<Image img={require('../../img/gcp_acc_1.png')} />

  1. Get the FULL EMAIL address of the corresponding Service Account

  2. Next, go to IAM & Admin > Manage Resources , select your top-level project that houses your Google Cloud Run Service

Click Add Principal

<Image img={require('../../img/gcp_acc_2.png')}/>

  1. Specify the Service Account as the principal and Vertex AI User as the role

<Image img={require('../../img/gcp_acc_3.png')}/>

Once that's done, when you deploy the new container in the Google Cloud Run service, LiteLLM will have automatic access to all Vertex AI endpoints.

s/o @Darien Kindlund for this tutorial