easy_fnc

The easy_fnc package provides a framework for generating responses using LLMs and executing user-defined functions. It allows users to define their own functions and integrate them with LLMs to create interactive and customizable applications. The package has optional dependencies for different LLM backends, such as Ollama and Groq, to provide flexibility in choosing the model that best suits the user's needs.

Beware that the package is still in development and may have breaking changes, as we have yet to release a stable version, easy_fnc 1.0.0.

Table of Contents

• Installation
• Usage
• Explanation of Different Modules
  • Defining User Functions
  • Core Utility Functions
  • Models
    • Ollama Model
    • Groq Model
  • Templates

Installation

To install the easy_fnc package, run the following command:

pip install easy_fnc

Usage

The full usage with Ollama is as follows (an example can also be found in usage.py in the root directory of the package):

from easy_fnc.function_caller import FunctionCallingEngine, create_functions_metadata
from easy_fnc.models import OllamaModel

# Declare the constants
MODEL_NAME = "adrienbrault/nous-hermes2pro-llama3-8b:f16"
VERBOSE = True

# Create a FunctionCallingEngine object
fnc_engine = FunctionCallingEngine()
fnc_engine.add_user_functions("easy_fnc/functions.py")
functions_metadata = create_functions_metadata(fnc_engine.functions)

# Create the Ollama model 

model = OllamaModel(
    MODEL_NAME, 
    functions_metadata
)

# Generate the response
user_input = "Can you get me a random city and the weather forecast for it?"
response_raw = model.generate(user_input)

# Print the raw response and system prompt if VERBOSE is True
if VERBOSE: 
    print(f"System prompt:\n {model.generate_system_prompt()}")
    print(f"Raw response:\n {response_raw}")

# Parse the example response
parsed_response = fnc_engine.parse_model_response(raw_response=response_raw)

# Print the parsed response
print(parsed_response)

# Call the functions
outputs = fnc_engine.call_functions(parsed_response.function_calls)

# Print the outputs
print(outputs)

Which would output, with the function call results displayed on the bottom of the output:

Thoughts:
The user wants to get a random city and its weather forecast. To do that, I will call the following functions:
1. get_random_city: This function retrieves a random city from a list.
2. get_weather_forecast: This function retrieves the weather forecast for a given location.

Function Calls:
- get_random_city:
  - kwargs: {}
  - returns: ['random_city']
- get_weather_forecast:
  - kwargs: {'location': 'random_city'}
  - returns: ['weather_forecast']

{'random_city': 'Rio de Janeiro', 'weather_forecast': {'location': 'Rio de Janeiro', 'forecast': 'rainy', 'temperature': '25°C'}}

Explanation of Different Modules

Defining User Functions

User-defined functions can be provided to the package as a .py file with the functions in it. A FunctionCallingEngine class is provided to facilitate the import and execution of user-defined functions, as well as the extraction of function calls from the model output.

Example:

# Create a FunctionCallingEngine object
fnc_engine = FunctionCallingEngine()
fnc_engine.add_user_functions("path/to/functions.py")
functions_metadata = create_functions_metadata(fnc_engine.functions)

Core Utility Functions

The package provides a set of core utility functions that can be used in conjunction with user-defined functions. These functions are defined in the easy_fnc/core_utils.py file and can be accessed using the get_core_utils function.

Example:

from easy_fnc.core_utils import get_core_utils

core_utils = get_core_utils()

Models

The package provides an abstract base class EasyFNCModel in the model.py file. This class defines the interface for implementing LLM interfaces that can generate responses based on user input and execute function calls.

To create a custom model, subclass EasyFNCModel and implement the following abstract method:

• generate(self, user_input: str) -> str: Generate a response based on the user input.

Example:

from easy_fnc.models.model import EasyFNCModel

class CustomModel(EasyFNCModel):
    def generate(self, user_input: str) -> str:
        # Implement response generation logic
        pass

Ollama Model

The package includes an implementation of the EasyFNCModel using the Ollama model. The OllamaModel class is defined in the easy_fnc/models/ollama.py file.

To use the Ollama model:

1. Pull the model from Ollama using the ollama pull command. For example:

ollama pull adrienbrault/nous-hermes2pro-llama3-8b:f16

2. Then, the usage is as follows if you're using the nous-hermes2pro-llama3-8b model:

from easy_fnc.models.ollama import OllamaModel
from easy_fnc.utils import load_template

# Create a FunctionCallingEngine object
fnc_engine = FunctionCallingEngine()
fnc_engine.add_user_functions("path/to/functions.py")
functions_metadata = create_functions_metadata(fnc_engine.functions)

# Create the Ollama model
MODEL_NAME = "adrienbrault/nous-hermes2pro-llama3-8b:f16"
ollama_model = OllamaModel(
    MODEL_NAME, 
    functions_metadata,
    template=load_template(file_type="toml")
)

Groq Model

The package also includes an implementation of the EasyFNCModel using the Groq model. The GroqModel class is defined in the easy_fnc/models/groq.py file.

To use the Groq model:

1. Add the environment variable GROQ_API_KEY with your Groq API key.
2. (Optional) Add the environment variable GROQ_MODEL with the model name you want to use. The default model is llama-3-70b-8192.
3. Then, the usage is as follows:

from easy_fnc.models import GroqModel

# Create a FunctionCallingEngine object
fnc_engine = FunctionCallingEngine()
fnc_engine.add_user_functions("path/to/functions.py")
functions_metadata = create_functions_metadata(fnc_engine.functions)

# Create the Groq model
model = GroqModel(functions_metadata)

Templates

The easy_fnc package uses JSON and TOML templates to format user input and model responses. The OllamaModel class accepts both a string and a dictionary as parameters for the template. The default template is defined in the easy_fnc/models/templates/base.toml file.

To use a custom template, you have two options:

1. Provide the template as a JSON file path using the load_template method:

from easy_fnc.utils import load_template

template = load_template(file_path="path/to/template.json", file_type="json")

2. Provide the template as a TOML file path using the load_template method:

from easy_fnc.utils import load_template

template = load_template(file_path="path/to/template.toml", file_type="toml")

Users have the flexibility to use either the default template or provide their own custom template to format the user input and model responses.

Contributing

Contributions to the easy_fnc package are welcome! If you find any issues or have suggestions for improvements, please open an issue or submit a pull request on the package's GitHub repository.

License

The easy_fnc package is open-source and released under the Apache 2.0 License.

This documentation provides an overview of how to use the easy_fnc package based on the provided files. It covers the key components, including user-defined functions, core utility functions, models, configuration, and templates. Users can refer to this documentation to understand how to integrate their own functions, create custom models, and utilize the package effectively in their applications.