Structured Output Response
Structured output is a feature that allows large language models (LLMs) to generate responses in a specific format with your resources. Structured output is best for extracting structured data to fit your application’s needs. It returns data formatted with the json_schema and json_schema_rules you provide.
To use the structured output feature you need to provide the following parameters:
Streaming is not supported when doing structured output response
| Parameter | Type | Description |
|---|
collection_id | String | The unique ID of the collection to query obtained from the /collection/create/ endpoint. |
request_query | String | The search term or query text. |
json_schema | String | The desired structure for the JSON response. |
json_schema_rules | String | Custom rules to guide the response’s adherence to the specified JSON schema. |
model(optional) | String | Specific model you desire to query response Check out our list of supported models here |
json_schema
The json_schema parameter defines the structure of the response you want the system to return. This parameter helps ensure the output matches your application’s requirements.
Example JSON Schema
{
"topic" : "string",
"description" : "string",
"count" : "int"
}
Example JSON Schema (List)
If you want a list of items and not just a single item, you can specify the json schema by putting the entire object in a list
[
{
"topic" : "string",
"description" : "string",
"count" : "int"
}
]
json_schema_rules
The json_schema_rules parameter allows you to define custom rules for how the schema is applied. Another way to think about this is that it is a set of instructions for the LLM to follow when generating the response. This can also be in form of a more detailed prompt This can include constraints on the data, additional formatting, or validation logic.
Example JSON Schema Rules
rules = "Give a very short description of every page on the document from"
Example JSON Schema Rules (List)
rules = ["Give a very short description of every page on the document from", "Make sure there are no less than 5 topics"]
Example Requests
from wetro import Wetrocloud
# Initialize the Wetrocloud client
client = Wetrocloud(api_key="your_api_key")
# Define a JSON schema for structured output
json_schema = [{"month": "", "sales": "int"}]
# Add processing rules
rules = "Give a breakdown of the sales for each month"
# Query with structured output requirements
response = client.collection.query_collection(
collection_id="my_unique_collection_id",
request_query="Generate a sales report",
json_schema=json_schema,
json_schema_rules=rules
)
print(response)
Example Response
{
"response": [
{
"month": "January",
"sales": 15000
},
{
"month": "February",
"sales": 18000
},
{
"month": "March",
"sales": 17000
}
],
"tokens": 4007,
"success": true
}
| Field | Description |
|---|
response | JSON object structured as per the provided schema. |
tokens | Number of tokens used for processing. |
success | Indicates whether the query was successful. |
