Function Calling

Function calling is a powerful capability that enables Large Language Models (LLMs) to interact with your code and external systems in a structured way. Instead of just generating text responses, LLMs can understand when to call specific functions and provide the necessary parameters to execute real-world actions.

How Function Calling Works

The process follows these steps:

Function Calling

This cycle can continue as needed, allowing for complex multi-step interactions between the application and the LLM.

Example Use Cases

Function calling is useful for many practical applications, such as:

  1. Data Retrieval: Converting natural language queries into API calls to fetch data (e.g., “Show me my recent orders” triggers a database query)
  2. Action Execution: Transforming user requests into specific function calls (e.g., “Schedule a meeting” becomes a calendar API call)
  3. Computation Tasks: Handling mathematical or logical operations through dedicated functions (e.g., calculating compound interest or statistical analysis)
  4. Data Processing Pipelines: Chaining multiple function calls together (e.g., fetching data → parsing → transformation → storage)
  5. UI/UX Integration: Triggering interface updates based on user interactions (e.g., updating map markers or displaying charts)

Using Tools (Function Definitions)

Tools are the primary way to define callable functions for your LLM. Each tool requires:

Here’s an example that defines weather-related functions:

from huggingface_hub import InferenceClient

client = InferenceClient("http://localhost:8080") # Replace with your HUGS host
messages = [
    {
        "role": "system",
        "content": "Don't make assumptions about values. Ask for clarification if needed.",
    },
    {
        "role": "user",
        "content": "What's the weather like the next 3 days in San Francisco, CA?",
    },
]

tools = [
    {
        "type": "function",
        "function": {
            "name": "get_n_day_weather_forecast",
            "description": "Get an N-day weather forecast",
            "parameters": {
                "type": "object",
                "properties": {
                    "location": {
                        "type": "string",
                        "description": "The city and state, e.g. San Francisco, CA",
                    },
                    "format": {
                        "type": "string",
                        "enum": ["celsius", "fahrenheit"],
                        "description": "The temperature unit to use",
                    },
                    "num_days": {
                        "type": "integer",
                        "description": "The number of days to forecast",
                    },
                },
                "required": ["location", "format", "num_days"],
            },
        },
    }
]

response = client.chat_completion(
    messages=messages,
    tools=tools,
    tool_choice="auto",
    max_tokens=500,
)
print(response.choices[0].message.tool_calls[0].function)
# ChatCompletionOutputFunctionDefinition(arguments={'format': 'celsius', 'location': 'San Francisco, CA', 'num_days': 3}, name='get_n_day_weather_forecast', description=None)

The model will analyze the user’s request and generate a structured call to the appropriate function with the correct parameters.

Using Pydantic Models for structured outputs

For better type safety and validation, you can use Pydantic models to define your function schemas. This approach provides:

Here’s how to use Pydantic models for function calling:

from pydantic import BaseModel, Field
from typing import List


class ParkObservation(BaseModel):
    location: str = Field(..., description="Where the observation took place")
    activity: str = Field(..., description="What activity was being done")
    animals_seen: int = Field(..., description="Number of animals spotted", ge=1, le=5)
    animals: List[str] = Field(..., description="List of animals observed")


client = InferenceClient("http://localhost:8080")  # Replace with your HUGS host
response_format = {"type": "json", "value": ParkObservation.model_json_schema()}


messages = [
    {
        "role": "user",
        "content": "I saw a puppy, a cat and a raccoon during my bike ride in the park.",
    },
]


response = client.chat_completion(
    messages=messages,
    response_format=response_format,
    max_tokens=500,
)
print(response.choices[0].message.content)
# {   "activity": "bike ride",
#     "animals": ["puppy", "cat", "raccoon"],
#     "animals_seen": 3,
#     "location": "the park"
# }

This will return a JSON object that matches your schema, making it easy to parse and use in your application.

Advanced Usage Patterns

Chaining Function Calls

LLMs can orchestrate multiple function calls to complete complex tasks:

tools = [
    {
        "type": "function",
        "function": {
            "name": "search_products",
            "description": "Search product catalog",
            "parameters": {
                "type": "object",
                "properties": {
                    "query": {"type": "string"},
                    "category": {"type": "string", "enum": ["electronics", "clothing", "books"]}
                }
            }
        }
    },
    {
        "type": "function",
        "function": {
            "name": "create_order",
            "description": "Create a new order",
            "parameters": {
                "type": "object",
                "properties": {
                    "product_id": {"type": "string"},
                    "quantity": {"type": "integer", "minimum": 1}
                }
            }
        }
    }
]

Error Handling and Execution

Always validate function calls before execution:

import json

def get_n_day_weather_forecast(location, format, num_days):
    return '{"temperature": 70, "condition": "sunny"}'

def handle_tool_call(tool_call):
    try:
        args = tool_call.function.arguments
        # Validate required parameters
        if tool_call.function.name == "get_n_day_weather_forecast":
            if not all(k in args for k in ["location", "format", "num_days"]):
                raise ValueError("Missing required parameters")
            # Only pass arguments that match the function's parameters
            valid_args = {k: v for k, v in args.items() 
                         if k in get_n_day_weather_forecast.__code__.co_varnames}
            return get_n_day_weather_forecast(**valid_args)
    except json.JSONDecodeError:
        return {"error": "Invalid function arguments"}
    except Exception as e:
        return {"error": str(e)}

res = handle_tool_call(response.choices[0].message.tool_calls[0])
print(res)
# {"temperature": 70, "condition": "sunny"}

Best Practices

  1. Function Design

  2. Error Handling

  3. Security

Never expose sensitive operations directly through function calls. Always implement proper validation and authorization checks.

For more information about basic inference capabilities, see our Inference Guide.

Update on GitHub