OpenAI Python API Library Wrapper

This wrapper provides the ability to stream sync and async responses when using or not function calling, it also provides default responses, memory manipulation and some debugging capabilities.

Installation

# install form PyPI
pip install openai-tools-gpt

Usage

Prior to interact with the Class, remember to setup your OpenAI Api Key in your environment.

Jupyter Notebook

os.environ["OPENAI_API_KEY"] = "Your_api_key",

Or in your terminal

export OPENAI_API_KEY="Your_api_key"

The model is set to "gpt-3.5-turbo" by default, full docstring can be inspect within the Class instantiation.

Simple use case

from openai_tools_gpt import ToolsGPT

gpt: ToolsGPT = ToolsGPT()

input: str = "Hey, what's up"

# Regular response
gpt.invoke(input)

# Synchronous streaming response
for token in gpt.stream(input):
    print(token, end="", flush=True)

# Asynchronous streaming response
async for token in gpt.astream(input):
    print(token, end="", flush=True)

Tools enabled

from typing import Callable
from openai_tools_gpt import ToolsGPT

tools_schema: dict = [
    {
                "type": "function",
                "function": {
                    "name": "get_dollar",
                    "description": "describe what is the function goal",
                    "parameters": {
                        "type": "object",
                        "properties": {},
                    },
                },
            }
]

def get_dollar() -> str:
    return "4000 COP"

tools: dict[str, Callable] = {"get_dollar": get_dollar}

gpt = ToolsGPT(tools_schema=tools_schema,
               tools=tools,
               tool_choice="auto")

input: str = "How's the dollar doing ?"

# Regular response
gpt.invoke(input)

# Synchronous streaming response
for token in gpt.stream(input):
    print(token, end="", flush=True)

# Asynchronous streaming response
async for token in gpt.astream(input):
    print(token, end="", flush=True)

Memory

The model has the memory set to False by default, so it won't remember any interaction, only the initial message (system or user) it set.

from openai_tools_gpt import ToolsGPT

gpt: ToolsGPT = ToolsGPT(with_memory=True,
                         memory=[{"role": "system",
                                  "content": "You are my friendly assistant"}])

Debugging

You can also see which tool or tools the model is using for generating the response, if using or not, for every single response.

from openai_tools_gpt import ToolsGPT

gpt: ToolsGPT = ToolsGPT(debug=True)

Useful links

• OpenAI Docs
• JSON Schema