Controlled JSON Generation

Ending Parsing Errors

When building applications, you often need the LLM to output structured data (like JSON) rather than plain text. Gemini supports response_schema.

from vertexai.generative_models import GenerativeModel, ResponseSchema, Type

schema = ResponseSchema(
    type=Type.OBJECT,
    properties={
        "recipe_name": ResponseSchema(type=Type.STRING),
        "ingredients": ResponseSchema(
            type=Type.ARRAY,
            items=ResponseSchema(type=Type.STRING)
        ),
    },
    required=["recipe_name", "ingredients"]
)

response = model.generate_content(
    "Give me a recipe for pancakes.",
    generation_config={"response_mime_type": "application/json", "response_schema": schema}
)