OpenAI tokens calculator, with function calls, images, and messages in one call
This package provides a utility function to estimate the token count for OpenAI chat completions.
To install the package, run the following command:
npm install openai-tokens-count
Here's an example of how to use the estimateTokens
function:
import { estimateTokens } from 'openai-tokens-count';
import OpenAI from "openai"; // for typings
const message: OpenAI.Chat.ChatCompletionCreateParamsNonStreaming = {
model: 'gpt-4-turbo',
messages: [{ role: 'user', content: 'Hello' }],
};
const run = async () => {
const estimatedTokens = await estimateTokens(message);
console.log('Estimated tokens:', estimatedTokens);
}
run();
The function returns the estimated token count for the given input.
For a more complex scenario, including multiple messages, tool calls, various parameters, and image estimation, you can use the following example:
import { estimateTokens } from 'openai-tokens-count';
import OpenAI from "openai"; // for typings
const advancedMessage: OpenAI.Chat.ChatCompletionCreateParamsNonStreaming = {
model: "gpt-4-turbo",
messages: [
{ role: "system", content: "You are a weather predictor" },
{ role: "user", content: "Hello! How cloudy is it in London?" },
{
role: "assistant",
content: "",
tool_calls: [
{
id: "call_w3cN5nYrqIbu6HLm7tYMP2OZ",
type: "function",
function: {
name: "get_current_weather_by_coords",
arguments: `{
"coords": {
"lat": "51.5074",
"long": "-0.1278"
},
"unit": "celsius"
}`
}
}
]
},
{
role: "tool",
content: '{ "temperature": 15, "condition": "cloudy" }',
tool_call_id: "call_w3cN5nYrqIbu6HLm7tYMP2OZ",
},
{
role: "user",
content: [
{
type: "text",
text: "What’s in this image?"
},
{
type: "image_url",
image_url: {
url: "https://raw.githubusercontent.com/n0isy/openai-tokens-count/master/tests/__fixtures__/1t-512x512.png",
detail: "high"
}
}
],
}
],
tools: [
{
type: "function",
function: {
name: "get_current_weather_by_coords",
description: "Get the current weather by coordinates",
parameters: {
type: "object",
properties: {
coords: {
type: "object",
description: "(lat, long)",
properties: {
lat: { type: "string", description: "latitude" },
long: { type: "string", description: "longitude" },
},
},
unit: { type: "string", enum: ["celsius", "fahrenheit"] },
},
required: ["coords"],
},
},
}
],
};
const runAdvanced = async () => {
const estimatedTokens = await estimateTokens(advancedMessage);
console.log('Estimated tokens for advanced message:', estimatedTokens);
}
runAdvanced();
To run the tests, use the following command:
npm test
The tests are defined in the test/run.spec.ts
file. They use the Jest testing framework to run test cases and compare the estimated token counts with the expected values.
To prepare test cases and generate the tokens.json
file, follow these steps:
-
Create a directory named
test/cases
and add test case files inside it. Each test case file should export an object with the following properties:model
(string): The name of the OpenAI model to use for chat completion.messages
(array): An array of message objects, each containing arole
andcontent
property.
Example test case file (
hello-world.ts
):export default { model: 'gpt-4-turbo', messages: [{ role: 'user', content: 'Hello' }], };
-
Run the
tests/prepare-from-openai.ts
script to generate thetokens.json
file:npm run makeTests
This script reads all the test case files from the
test/cases
directory, sends them to the OpenAI API for chat completion, and saves the actual token counts in thetest/tokens.json
file. -
The
tokens.json
file will be used by the tests to compare the estimated token counts with the actual values.Example
tokens.json
file:{ "hello-world.ts": 8 }
Now you're ready to run the tests and verify the token estimation functionality.
Contributions are welcome! If you find any issues or have suggestions for improvements, please open an issue or submit a pull request.
This package is open-source and available under the MIT License.