Estimate OpenAI token usage for chat completions, including functions, with this Python utility!
This package is based upon hmarr's openai-chat-tokens. As of right now (September 2023) there is no official documentation from openai on how to accurately predict the number of tokens from functions. This package solves that! Use it to get a very precise estimation of the token count for chat completions and better manage your OpenAI API usage.
Most often it is correct down to the token.
-
Install the Package via pip
pip install openai-function-tokens -
Import the Estimation Function
from openai_function_tokens import estimate_tokens
To use the estimator, call the estimate_tokens function:
estimate_tokens(messages, functions=None, function_call=None)Pass in the messages, and optionally functions and function_call, to receive a precise token count.
Credit to hmarr for the original TypeScript tool. For a better understanding of token counting logic, check out his blog post.
How to call functions with chat models
How to use functions with a knowledge base
Counting tokens (only messages)
Feedback, suggestions, and contributions are highly appreciated. Help make this tool even better!