Skip to content
/ ezgpt Public

Python library for easier GPT usage than the default openai library.

License

Notifications You must be signed in to change notification settings

Ascyt/ezgpt

Repository files navigation

ezgpt

ezgpt is a Python library designed to simplify the interaction with OpenAI's GPT (Generative Pre-trained Transformer) models. It provides a convenient interface for sending prompts to the model and receiving responses, with optional logging for debugging purposes.

Installation

You can install ezgpt via pip:

pip install ezgpt

Usage

To use ezgpt, you need to have an API key from OpenAI. This key must either be set as an environment variable OPENAI_API_KEY or be passed with the static function ezgpt.set_api_key(api_key) before using the library. Since these are async functions, you either have to use await, or import asyncio and use asyncio.run().

Initialization

First, import the ezgpt module and initialize the gpt class:

import ezgpt

gpt = ezgpt.gpt()

The gpt class constructor accepts the following parameters:

  • model: The identifier of the GPT model to use (default: 'gpt-3.5-turbo').
  • system: An optional system-level prompt that provides instructions or context for the GPT model.
  • temperature: Controls randomness in the response generation (default: 0).
  • top_p: Controls diversity of the response generation (default: 0).
  • max_tokens: The maximum number of tokens to generate in the response (default: 2048).
  • frequency_penalty: Decreases the likelihood of repetition in the response (default: 0).
  • presence_penalty: Encourages the model to talk about new topics (default: 0).
  • logs: Enables or disables logging of the interaction (default: False).

Sending Prompts

To send a prompt to the GPT model, use the get method of the gpt instance:

response = await gpt.get(user="Your prompt here")

The get method accepts the following parameters:

  • system: Overrides the system-level prompt for this request.
  • user: The user-level prompt to send to the model.
  • messages: A list of previous message exchanges to maintain context.
  • temperature: Overrides the default temperature for this request.
  • top_p: Overrides the default top_p for this request.
  • max_tokens: Overrides the default max_tokens for this request.
  • frequency_penalty: Overrides the default frequency_penalty for this request.
  • presence_penalty: Overrides the default presence_penalty for this request.

Since your last request is stored under self.previous, you can append a message to your conversation like that:

response = await gpt.get(messages=gpt.previous, user="Another message")

Logging

If logging is enabled, ezgpt will print the interaction with the GPT model to the console. This includes the prompts sent by the user and system, as well as the responses from the assistant.

Simpler usage

If you don't care about any classes or instances, you can also just use the static get and reset functions:

await ezgpt.get('Hello!') 
# 'Hello! How can I assist you today?'

Aside the normal GPT arguments, there's another boolean use_previous to use the previous conversation:

await ezgpt.get('What was my previous message?', True) 
# 'Your previous message was "Hello".'

You can use the reset() function to clear the previous messages, and optionally also change the model:

ezgpt.reset('gpt-4')

Conversation

An even simpler way to use GPT is using the conversation() static function. Here you can use the normal GPT arguments, and when you use it, you'll be able to have a conversation with GPT.

There are special commands you can use:

  • ?: View the list of commands
  • ??: Print some general useful information
  • !: Exit the conversation
  • :[command]: Run Python command: This will execute everything after the : as a line of Python.
    • Example: :print('hello') will print hello
  • #[property] [value]: Set GPT's property, such as model. Use just # to list all properties. Use ## to reset all properties. Use ?# for help.
    • Example: #model gpt-4 will set the model to gpt-4
  • $[message]: Set the system message. Leave message empty to remove. Use _ as message for multiline
  • +<index> [message]: Insert message before index (double + for assistant instead of user). Index defaults to last message (append). Use _ as message for multiline
    • Example: ++0 Hello! will insert a message at the start by the assistant with content Hello!
  • -<index>: Remove message at index (double - to clear every message, triple - to completely restart ezgpt.conversation). Index defaults to last message
    • Example: -0 will remove the first message
  • ~[index] [message]: Change message at index (double ~ for reverse role). If you use ~~ you don't need a message argument. Use _ as message for multiline
    • Example: ~~0 Hello! will change the first message to Hello! and switch its role
  • &: Re-generates the last message GPT sent.
  • @<index>: Copy message at index (defaults to last message) to clipboard (requires pyperclip module)
  • @~<index>: Copy codeblock. Index starts and defaults to 1 (first codeblock).
  • @@: Copy conversation JSON to clipboard (requires pyperclip module)
  • ^<name>: Save the conversation. Gets cleared after restart and is only valid for the current session. arg defaults to the current conversation
  • %<name>: Load the conversation name. Use just % without arguments to list all saved conversations. Use %%<name> to use from local filesystem (~/.ezgpt_conversations.json)
  • %+<name>: Save the conversation as name. Use %%+<name> to save to local filesystem.
  • %-<name>: Remove the conversation name. Use %%
  • %%<name> Same as %<name> but will get the conversation from ~/.ezgpt_conversations.json
  • Flows: Similar syntax to conversation saving/loading, but using ^ instead of %. To run a flow, use ^[name] [message]. Use _ as [message] for multi-line. To save, use ^+[name] [conversations], where conversations is a list of saved conversations to step through, seperated by spaces. Also allows for local filesystem (~/.ezgpt_flows.json)
  • =: Switch between full view and shortened view. Shortened view will only contain the first line up to 100 characters for easy readability.
  • empty: Sending an empty message will reload the conversation to only include the actual messages.
  • \[message]: This will let you type messages with special characters at the beginning without having them act as commands.
    • Example: \- Hello will add message - Hello
  • _: Start multiline. This lets you write and paste in multi-line text. Ctrl+X with Enter to stop multiline, Ctrl+C to cancel, Ctrl+U with Enter to remove previous line.

If you put a number before your prompt, it will execute the prompt n times.

  • Example: 10-0 will remove the first 10 messages

A shorthand to start conversation is using the ezgpt.c() static function. While generating, you can use Ctrl+C to cancel or resend.

Notes

  • The ezgpt library assumes that the OpenAI API key is set in the environment variable OPENAI_API_KEY or has been passed using ezgpt.set_api_key(api_key).
  • The gpt class maintains a history of the conversation in the previous attribute, which can be used to provide context for subsequent requests.

About

Python library for easier GPT usage than the default openai library.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published