Vertex Generative AI SDK for Python

The Vertex Generative AI SDK helps developers use Google’s generative AI Gemini models to build AI-powered features and applications. The SDKs support use cases like the following:

  • Generate text from texts, images and videos (multimodal generation)

  • Build stateful multi-turn conversations (chat)

  • Function calling

Installation

To install the google-cloud-aiplatform Python package, run the following command:

pip3 install --upgrade --user "google-cloud-aiplatform>=1.38"

Usage

For detailed instructions, see quickstart and Introduction to multimodal classes in the Vertex AI SDK.

Imports:

import vertexai

Initialization:

vertexai.init(project='my-project', location='us-central1')

Basic generation:

from vertexai.generative_models import GenerativeModel
model = GenerativeModel("gemini-pro")
print(model.generate_content("Why is sky blue?"))

Using images and videos

from vertexai.generative_models import GenerativeModel, Image
vision_model = GenerativeModel("gemini-pro-vision")

# Local image
image = Image.load_from_file("image.jpg")
print(vision_model.generate_content(["What is shown in this image?", image]))

# Image from Cloud Storage
image_part = generative_models.Part.from_uri("gs://download.tensorflow.org/example_images/320px-Felis_catus-cat_on_snow.jpg", mime_type="image/jpeg")
print(vision_model.generate_content([image_part, "Describe this image?"]))

# Text and video
video_part = Part.from_uri("gs://cloud-samples-data/video/animals.mp4", mime_type="video/mp4")
print(vision_model.generate_content(["What is in the video? ", video_part]))

Chat

from vertexai.generative_models import GenerativeModel, Image
vision_model = GenerativeModel("gemini-ultra-vision")
vision_chat = vision_model.start_chat()
image = Image.load_from_file("image.jpg")
print(vision_chat.send_message(["I like this image.", image]))
print(vision_chat.send_message("What things do I like?."))

System instructions

from vertexai.generative_models import GenerativeModel
model = GenerativeModel(
    "gemini-1.0-pro",
    system_instruction=[
        "Talk like a pirate.",
        "Don't use rude words.",
    ],
)
print(model.generate_content("Why is sky blue?"))

Function calling

# First, create tools that the model is can use to answer your questions.
# Describe a function by specifying it's schema (JsonSchema format)
get_current_weather_func = generative_models.FunctionDeclaration(
    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"
        ]
    },
)
# Tool is a collection of related functions
weather_tool = generative_models.Tool(
    function_declarations=[get_current_weather_func],
)

# Use tools in chat:
model = GenerativeModel(
    "gemini-pro",
    # You can specify tools when creating a model to avoid having to send them with every request.
    tools=[weather_tool],
)
chat = model.start_chat()
# Send a message to the model. The model will respond with a function call.
print(chat.send_message("What is the weather like in Boston?"))
# Then send a function response to the model. The model will use it to answer.
print(chat.send_message(
    Part.from_function_response(
        name="get_current_weather",
        response={
            "content": {"weather": "super nice"},
        }
    ),
))

Automatic Function calling

Note: The FunctionDeclaration.from_func converter does not support nested types for parameters. Please provide full FunctionDeclaration instead.

from vertexai.preview.generative_models import GenerativeModel, Tool, FunctionDeclaration, AutomaticFunctionCallingResponder

# First, create functions that the model can use to answer your questions.
def get_current_weather(location: str, unit: str = "centigrade"):
    """Gets weather in the specified location.

    Args:
        location: The location for which to get the weather.
        unit: Optional. Temperature unit. Can be Centigrade or Fahrenheit. Defaults to Centigrade.
    """
    return dict(
        location=location,
        unit=unit,
        weather="Super nice, but maybe a bit hot.",
    )

# Infer function schema
get_current_weather_func = FunctionDeclaration.from_func(get_current_weather)
# Tool is a collection of related functions
weather_tool = Tool(
    function_declarations=[get_current_weather_func],
)

# Use tools in chat:
model = GenerativeModel(
    "gemini-pro",
    # You can specify tools when creating a model to avoid having to send them with every request.
    tools=[weather_tool],
)

# Activate automatic function calling:
afc_responder = AutomaticFunctionCallingResponder(
    # Optional:
    max_automatic_function_calls=5,
)
chat = model.start_chat(responder=afc_responder)
# Send a message to the model. The model will respond with a function call.
# The SDK will automatically call the requested function and respond to the model.
# The model will use the function call response to answer the original question.
print(chat.send_message("What is the weather like in Boston?"))

Evaluation

  • To perform bring-your-own-response(BYOR) evaluation, provide the model responses in the response column in the dataset. If a pairwise metric is used for BYOR evaluation, provide the baseline model responses in the baseline_model_response column.
import pandas as pd
from vertexai.evaluation import EvalTask, MetricPromptTemplateExamples

eval_dataset = pd.DataFrame({
        "prompt"  : [...],
        "reference": [...],
        "response" : [...],
        "baseline_model_response": [...],
})
eval_task = EvalTask(
    dataset=eval_dataset,
    metrics=[
            "bleu",
            "rouge_l_sum",
            MetricPromptTemplateExamples.Pointwise.FLUENCY,
            MetricPromptTemplateExamples.Pairwise.SAFETY
    ],
    experiment="my-experiment",
)
eval_result = eval_task.evaluate(experiment_run_name="eval-experiment-run")
  • To perform evaluation with Gemini model inference, specify the model parameter with a GenerativeModel instance. The input column name to the model is prompt and must be present in the dataset.
from vertexai.evaluation import EvalTask
from vertexai.generative_models import GenerativeModel

eval_dataset = pd.DataFrame({
    "reference": [...],
    "prompt"  : [...],
})
result = EvalTask(
    dataset=eval_dataset,
    metrics=["exact_match", "bleu", "rouge_1", "rouge_l_sum"],
    experiment="my-experiment",
).evaluate(
    model=GenerativeModel("gemini-1.5-pro"),
    experiment_run_name="gemini-eval-run"
)
  • If a prompt_template is specified, the prompt column is not required. Prompts can be assembled from the evaluation dataset, and all prompt template variable names must be present in the dataset columns.
import pandas as pd
from vertexai.evaluation import EvalTask, MetricPromptTemplateExamples
from vertexai.generative_models import GenerativeModel

eval_dataset = pd.DataFrame({
    "context"    : [...],
    "instruction": [...],
})
result = EvalTask(
    dataset=eval_dataset,
    metrics=[MetricPromptTemplateExamples.Pointwise.SUMMARIZATION_QUALITY],
).evaluate(
    model=GenerativeModel("gemini-1.5-pro"),
    prompt_template="{instruction}. Article: {context}. Summary:",
)
  • To perform evaluation with custom model inference, specify the model parameter with a custom inference function. The input column name to the custom inference function is prompt and must be present in the dataset.
from openai import OpenAI
from vertexai.evaluation import EvalTask, MetricPromptTemplateExamples


client = OpenAI()
def custom_model_fn(input: str) -> str:
    response = client.chat.completions.create(
        model="gpt-3.5-turbo",
        messages=[
        {"role": "user", "content": input}
        ]
    )
    return response.choices[0].message.content

eval_dataset = pd.DataFrame({
    "prompt"  : [...],
    "reference": [...],
})
result = EvalTask(
    dataset=eval_dataset,
    metrics=[MetricPromptTemplateExamples.Pointwise.SAFETY],
    experiment="my-experiment",
).evaluate(
    model=custom_model_fn,
    experiment_run_name="gpt-eval-run"
)
  • To perform pairwise metric evaluation with model inference step, specify the baseline_model input to a PairwiseMetric instance and the candidate model input to the EvalTask.evaluate() function. The input column name to both models is prompt and must be present in the dataset.
import pandas as pd
from vertexai.evaluation import EvalTask, MetricPromptTemplateExamples, PairwiseMetric
from vertexai.generative_models import GenerativeModel

baseline_model = GenerativeModel("gemini-1.0-pro")
candidate_model = GenerativeModel("gemini-1.5-pro")

pairwise_groundedness = PairwiseMetric(
    metric_prompt_template=MetricPromptTemplateExamples.get_prompt_template(
        "pairwise_groundedness"
    ),
    baseline_model=baseline_model,
)
eval_dataset = pd.DataFrame({
    "prompt"  : [...],
})
result = EvalTask(
    dataset=eval_dataset,
    metrics=[pairwise_groundedness],
    experiment="my-pairwise-experiment",
).evaluate(
    model=candidate_model,
    experiment_run_name="gemini-pairwise-eval-run",
)

Documentation

You can find complete documentation for the Vertex AI SDKs and the Gemini model in the Google Cloud documentation

Contributing

See Contributing for more information on contributing to the Vertex AI Python SDK.

License

The contents of this repository are licensed under the Apache License, version 2.0.