EasyPrompting

EasyPrompting is a Python library for simplifying language model (LM) interactions.

Features

Prompting LMs usually entails a couple of challenges:

• Supporting different kinds of LM APIs
• Logging and debugging interactions
• Caching LM responses to save on time and cost
• Having persistant memory over longer interactions
• Getting responses in a specified format from the LM

EasyPrompting is designed to tacle these challenges.

Demo

Here is how to make a simple chat bot using the logger and debugger features:

from easy_prompting.prebuilt import GPT, Prompter, PrintLogger, PrintDebugger

def chat_bot() -> None:
    """Chat with an LM"""
    lm = GPT("gpt-4o-mini")
    prompter = Prompter(lm)
    prompter.set_logger(PrintLogger()) # print conversation
    prompter.set_debugger(PrintDebugger()) # get user input via debug mode
    prompter.set_tag("chat bot") # set a conversation tag
    prompter.set_cache("completions") # set the cache directory
    prompter.add_message(
        "You are a ChatBot. Talk with the user.",
        role="developer"
    )
    while True:
        # 1. Wait for user input (debug mode)
        # 2. Get response from LM
        prompter.add_completion()

Shell output:

Message(tag='chat bot', role='developer', idx=0):
  You are a ChatBot. Talk with the user.

Input(role='user', continue='↵', exit='Ctrl+c'):
  Hi :) # Get user input

Message(tag='chat bot', role='user', idx=1):
  Hi :) # Send user input

Message(tag='chat bot', role='assistant', idx=2):
  Hello! 😊 How can I assist you today?

easy_prompting/__main__.py demonstrates simple use cases for this library.

Setup

SETUP.md contains all the setup instructions.

Manual

EasyPrompting provides two main entry points:

• easy_prompting which contains the core architecture
• easy_prompting.prebuilt which additionally contains auxilary architecture for convinience for common use cases

Core Architecture

The core architecture focuses on the prompter class. It handles the interaction with the LM, as well as the caching of LM respones.

The LM class provides an abstract interface to any given LM implementation.

The Logger class provides an abstract interface for logging and printing conversations between the prompter and the LM.

The Debugger class provides an abstract interface for interacting with the LM in conversations for debugging purposes.

The Instruction class provides an abstract interface for instructing the LM on how it should format responses, and how to extract relevant information from those responses.

Auxilary Architecture

The auxilary architecture mainly provides implementations of the core architecture interfaces for common use cases, as well as some utilities.