ChatGPTBlock

A Python package for interacting with OpenAI's chat models through the OpenAI API.

Table Of Contents

• Introduction
• Features
• Installation
• Usage
• Contributing
• License

Introduction

The main purpose of the package is to have a simple interface for interacting with OpenAI api. Both in streamable and non-streamable fashion. OpenAI API itself does not store the history and can throw and error if history is too long. The package provides the solution to this by simply counting the length of the history and trimming it when needed.

Features

1. streamable mode
2. supporting the history of the conversation
3. resetting the history of the conversation
4. adding your custom pre-processing

Installation

Install Locally

1. git clone https://github.com/SkuratovichA/chatgpt_block
2. cd chatgpt_block
3. pip install -e .

Install With PIP

pip install chatgpt_block

Usage

Creating The Conversation

Here's an example of how to use the ChatGPTBlock class: There are two options of using the class. Streamable and non-streamable.

Non-Streamable Mode

from chatgpt_block import ChatGPTBlock

# Initialize the ChatGPTBlock instance
chat_gpt_block = ChatGPTBlock(
    system_prompt="You are a helpful assistant.",
    openai_api_key="your_openai_api_key",
    model="gpt-4",
    preprocessor=lambda x: x,
)

# Get a response from the model
response = chat_gpt_block("Tell me a joke.")
print(response)

Streamable Mode

from chatgpt_block import ChatGPTBlock

chat_gpt_block = ChatGPTBlock(
    system_prompt="You are a helpful assistant.",
    openai_api_key="your_openai_api_key",
    model="gpt-4",
    stream=True,
    preprocessor=lambda x: x,
)

generator = chat_gpt_block("Tell me a joke.")
for token in generator:
    print(token, flush=True, end='')

Resetting The Conversation

from chatgpt_block import ChatGPTBlock

chat_gpt_block = ChatGPTBlock(
    system_prompt="You are a helpful assistant.",
    openai_api_key="your_openai_api_key",
    model="gpt-4",
    stream=True,
    preprocessor=lambda x: x,
)

chat_gpt_block("Tell me a joke.")
print(chat_gpt_block.answer)

chat_gpt_block.reset()

response = chat_gpt_block("Tell me a story.")
print(response)

Custom Preprocessor Function

from chatgpt_block import ChatGPTBlock

def custom_preprocessor(input_text: str, some_dictionary: dict, prologue: str) -> str:
    return f"{prologue}\n{input_text}. {some_dictionary}"

chat_gpt_block = ChatGPTBlock(
    system_prompt="You are a helpful assistant.",
    openai_api_key="your_openai_api_key",
    model="gpt-4",
    preprocessor=custom_preprocessor,
)

response = chat_gpt_block(
    input_text="please check whether this python dictionary is valid", 
    some_dictionary={"hello": "world"}, prologue=''
)
print(response)

Custom Error Handling Function

from chatgpt_block import ChatGPTBlock

def custom_error_handler():
    print("There was an error when communicating with OpenAI API. Please try again later")

chat_gpt_block = ChatGPTBlock(
    system_prompt="You are a helpful assistant.",
    openai_api_key="your_openai_api_key",
    model="gpt-4",
    on_error=custom_error_handler,
    preprocessor=lambda x: x,
)

response = chat_gpt_block("tell me a joke")
print(response)

Feeding The Model With The Examples

from chatgpt_block import ChatGPTBlock

examples = [
    ("Please add a vegetable to an animal", 
     "Alpaca + avocado = alpacavocado"),
    ("Now I want something with cats", 
     "Cat + Cappuccino = cappucicatto"),
]

chat_gpt_block = ChatGPTBlock(
    system_prompt=
    """You are a helpful assistant with a sense of humor. 
       You know everything about animals, vegetables and how to combine them. 
       You must add animals to vegetables no matter what to create funny words.
    """,
    examples=examples,
    openai_api_key="your_openai_api_key",
    model="gpt-4",
    preprocessor=lambda x: x,
)

response = chat_gpt_block("Make up some new word with \"tortoise\"")
print(response)

Silent Errors

from chatgpt_block import ChatGPTBlock

# Initialize the ChatGPTBlock instance
chat_gpt_block = ChatGPTBlock(
    system_prompt="wrong key",
    openai_api_key="your_openai_api_key",
    model="gpt-4",
    preprocessor=lambda x: x,
    raise_on_error=False,
)

# Get a response from the model
response = chat_gpt_block("Tell me a joke.")
print(response)  # OpenAI internal error. No API key provided. You can ...

Errors as Exceptions

from chatgpt_block import ChatGPTBlock

# Initialize the ChatGPTBlock instance
chat_gpt_block = ChatGPTBlock(
    system_prompt="wrong key",
    openai_api_key="your_openai_api_key",
    model="gpt-4",
    preprocessor=lambda x: x,
    raise_on_error=True,
)

# Get a response from the model
response = chat_gpt_block("Tell me a joke.")
# AuthenticationError: No API key provided ...

Contributing

This package is free to any ideas. Just create an issue or a pull request on GitHub.

License

chatgpt_block is released under MIT License.