llm-messages-token-helper

A helper library for estimating tokens used by messages and building messages lists that fit within the token limits of a model. Currently designed to work with the OpenAI GPT models (including GPT-4 turbo with vision). Uses the tiktoken library for tokenizing text and the Pillow library for image-related calculations.

Installation

Install the package:

python3 -m pip install llm-messages-token-helper

Usage

The library provides the following functions:

• build_messages
• count_tokens_for_message
• count_tokens_for_image
• get_token_limit

build_messages

Build a list of messages for a chat conversation, given the system prompt, new user message, and past messages. The function will truncate the history of past messages if necessary to stay within the token limit.

Arguments:

• model (str): The model name to use for token calculation, like gpt-3.5-turbo.
• system_prompt (str): The initial system prompt message.
• new_user_message (str | List[openai.types.chat.ChatCompletionContentPartParam]): The new user message to append.
• past_messages (list[dict]): The list of past messages in the conversation.
• few_shots (list[dict]): A few-shot list of messages to insert after the system prompt.
• max_tokens (int): The maximum number of tokens allowed for the conversation.

Returns:

• list[openai.types.chat.ChatCompletionMessageParam]

Example:

from llm_messages_token_helper import build_messages

messages = build_messages(
    model="gpt-35-turbo",
    system_prompt="You are a bot.",
    new_user_message="That wasn't a good poem.",
    past_messages=[
        {
            "role": "user",
            "content": "Write me a poem",
        },
        {
            "role": "assistant",
            "content": "Tuna tuna I love tuna",
        },
    ],
    few_shots=[
        {
            "role": "user",
            "content": "Write me a poem",
        },
        {
            "role": "assistant",
            "content": "Tuna tuna is the best",
        },
    ]
)

count_tokens_for_message

Counts the number of tokens in a message.

Arguments:

• model (str): The model name to use for token calculation, like gpt-3.5-turbo.

Returns:

• int: The number of tokens in the message.

Example:

from llm_messages_token_helper import count_tokens_for_message

message = {
    "role": "user",
    "content": "Hello, how are you?",
}
model = "gpt-4"
num_tokens = count_tokens_for_message(model, message)

count_tokens_for_image

Count the number of tokens for an image sent to GPT-4-vision, in base64 format.

Arguments:

• image (str): The base64-encoded image.

Returns:

• int: The number of tokens used up for the image.

Example:

Count the number of tokens for an image sent to GPT-4-vision:

```python
from llm_messages_token_helper import count_tokens_for_image

image = "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEA..."
num_tokens = count_tokens_for_image(image)

get_token_limit

Get the token limit for a given GPT model name (OpenAI.com or Azure OpenAI supported).

Arguments:

• model (str): The model name to use for token calculation, like gpt-3.5-turbo (OpenAI.com) or gpt-35-turbo (Azure).

Returns:

• int: The token limit for the model.

Example:

from llm_messages_token_helper import get_token_limit

model = "gpt-4"
max_tokens = get_token_limit(model)