TokenTrim

0.1.13 verified Mon Apr 27 auth: no python

A utility library for trimming 'messages' arrays to fit within a specified token limit, commonly used with GPT models. Current version is 0.1.13, released as a stable package with no recent changes.

pip install tokentrim

Common errors

error ModuleNotFoundError: No module named 'tiktoken' ↓

cause Missing dependency tiktoken.

fix

Install tiktoken: pip install tiktoken

error KeyError: 'model' ↓

cause The model string provided is not recognized by tiktoken.

fix

Use a valid model like 'gpt-3.5-turbo', 'gpt-4', or 'text-davinci-003'.

Warnings

gotcha If the last message is a user message and trimming removes it, the function may return an empty list or incomplete conversation. ↓

fix Ensure that the last message is preserved or handle the case where no messages remain.

gotcha The 'model' parameter must match a model supported by tiktoken; otherwise, it may fall back to a default encoding or raise an error. ↓

fix Use a known model string like 'gpt-3.5-turbo' or 'gpt-4'.

deprecated Future versions may change default trimming behavior; check documentation for updates. ↓

fix Always specify max_tokens explicitly to avoid breaking changes.

Imports

trim_messages
```
from tokentrim import trim_messages
```
Main function for trimming message arrays.

Quickstart

Basic usage: trim a list of messages to a maximum token count for a given model.

import tokentrim

messages = [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": "Tell me about Python."}
]

# Trim messages to fit within 100 tokens
trimmed_messages = tokentrim.trim_messages(messages, model="gpt-3.5-turbo", max_tokens=100)
print(trimmed_messages)