Function calling enhances the capabilities of Large Language Models (LLMs) in two main ways:
-
Structured Responses: LLMs can generate structured responses, like JSON objects, that can be used as arguments in subsequent functions within LLM applications. This is a more secured way to ensure that your application is in control of executing a set of functions for which the LLM has generated function arguments.
-
Function Invocation: Optionally, LLMs can directly invoke functions using these structured JSON objects as arguments. This allows them to utilize external tools such as Python interpreters, search websites, or access external databases, offering vast potention of possibilities.
However, direct function invocation by LLMs should be used cautiously, as it can inadvertly call a funciton with generated arguments that you may not wish to invoke automatically (with undesirable consequences). For examplem, an LLM-generated e-mail response or deletion of a record response from an enterprise data source.
Use this extended and automatic LLM invocation of your desired functions judiciously. Nonetheless, both the aforementioned features accord LLMs with extended functionality to build AI agents for automatic tasks.
To that extent, notebooks and examples in this section demonstrate:
- Creating or generating structured JSON for a downstream Python function within an LLM application
- Having the LLM invoke a Python function with JSON arguments and provide the results
- Generating SQL queries by asking LLM using natural language. These queries are then executed, and its results fed back into LLM to generate the final response to the user
- Using parallel function calling functionality using compatile OpenAI APIs
Explore the vast possibilities of function calling with OpenAI and Anyscale Endpoints.
| Notebook Description | Open with Colab |
|---|---|
| How to generate function arguments for OpenAI |  |
| How to generate function arguments for Anyscale Enpoints | |
| Parallel function calling for OpenAI and Anyscale Endpoints | |
Other Python applications showcasing the same as the notebooks above:
| Python file Description | View it on Github |
|---|---|
| Open AI and Anyscale Endpoints parallel function calling an external SQLite | Python App |
| Open AI and Anyscale Endpoints parallel function calling an external Weather API | Python App |