Interact with Google’s Vertex AI Text Generation Models

A minimal example demonstrating how to use Google Cloud text generation models via the python SDK

Author
Published

December 7, 2023

This provides a minimal example of how to interact with Google’s Text-Bison model – a text generation model offered through GCP’s VertexAI suite of tools.

#if aiplatform isn't installed, do that
pip install --upgrade google-cloud-aiplatform

This will show how to pass a prompt and some hyperparameters to the text generation model, then view a response passed back.

import vertexai
from vertexai.languagemodels import TextGenerationModel

#you might need to initiate a project
vertexai.init(project="my-project-id")

#instantiate a model
model = TextGenerationModel.from_pretrained("text-bison@001")

In the above, "text-bison@001" is the model we’re passing our prompt to. This is an ok default, but we might want to see the list of available models to choose a different one. This is especiall true if we need to pass a large piece of text to a model.

#p is our prompt
p = "Generate 10 interview questions that would be suitable to ask a candidate for an educational researcher position"

#obviously we can tweak these as needed
#but these feel like decent defaults
params = {
    "temperature": 0.2,
    "top_p": 0.9,
    "top_k": 40,
    "max_output_tokens": 1024
}

#hit the model and get a response
resp = model.predict(p, **params)

#view the text from the response
print(resp.text)