How to use Functions with Multiple Arguments?

[vtong]

NVIDIA Agent Intelligence toolkit

Functions with Multiple Arguments

I have a question about Functions with Multiple Arguments. I defined function similar to this. But I have not idea. How to set the params in config.yml. I have no idea on the prompt and syntax. Can anyone teach me how to use Multiple Arguments?

Or does anyone know which examples/ has the sample code for my reference?

Thanks you so much.

[slopp]

Hi @vtong - one way to do this is to pass the input as JSON, eg

aiq run --config_file … --input ‘{“input1”: “value1”, “input2”:“value2”}’

That input will then automatically be converted to the appropriate multi-argument input for your function. See Writing Custom Functions — NVIDIA Agent Intelligence Toolkit (1.1.0)

If you want more control over how the input is parsed into your function arguments you can create a converter, here is an example:

import logging

from pydantic import Field

from aiq.builder.builder import Builder
from aiq.builder.function_info import FunctionInfo
from aiq.cli.register_workflow import register_function
from aiq.data_models.function import FunctionBaseConfig
from pydantic import BaseModel

logger = logging.getLogger(__name__)


class RegistryExampleFunctionConfig(FunctionBaseConfig, name="registry_example"):
    """
    AIQ Toolkit function template. Please update the description.
    """
    # Add your custom configuration parameters here
    parameter: str = Field(default="default_value", description="Notional description for this parameter")

class MyFunctionInput(BaseModel):
   input1: str
   input2: str

@register_function(config_type=RegistryExampleFunctionConfig)
async def registry_example_function(
    config: RegistryExampleFunctionConfig, builder: Builder
):
    # Implement your function logic here
    async def _response_fn(input_message: MyFunctionInput) -> str:
        # Process the input_message and generate output
        output_message = f"{input_message.input1} and {input_message.input2}"
        return output_message
    
    def converter(input_message: str) -> MyFunctionInput:
        return MyFunctionInput.model_validate_json(input_message)
        

    try:
        yield FunctionInfo.create(single_fn=_response_fn, converters=[converter])
    except GeneratorExit:
        print("Function exited early!")
    finally:
        print("Cleaning up registry_example workflow.")

For more info, Writing Custom Functions — NVIDIA Agent Intelligence Toolkit (1.1.0)

[vtong]

Great! The key point is “–input ‘{“input1”: “value1”, “input2”:“value2”}’” 👍