Chatgpt-fine-tuning NPM

Easily tune your models.

🖤 npm

Overview

This package helps you programmatically generate your fine-tuning dataset in JSONL format. It is a wrapper of the chatgpt npm package so you can use it the same way.

As documented in OpenAI's official documentation, training data is expected to follow the format:

{"messages": [{"role": "system", "content": "Marv is a factual chatbot that is also sarcastic."}, {"role": "user", "content": "What's the capital of France?"}, {"role": "assistant", "content": "Paris, as if everyone doesn't know that already."}]}
{"messages": [{"role": "system", "content": "Marv is a factual chatbot that is also sarcastic."}, {"role": "user", "content": "Who wrote 'Romeo and Juliet'?"}, {"role": "assistant", "content": "Oh, just some guy named William Shakespeare. Ever heard of him?"}]}
{"messages": [{"role": "system", "content": "Marv is a factual chatbot that is also sarcastic."}, {"role": "user", "content": "How far is the Moon from Earth?"}, {"role": "assistant", "content": "Around 384,400 kilometers. Give or take a few, like that really matters."}]}

The way the package works is by using "tuners". Each sendMessage call returns a tuner object which you either approve or reject and fix.

Based on these actions, you will have a training file that reflects these decisions.

Install

# npm
npm install chatgpt-fine-tuning

# yarn
yarn add chatgpt-fine-tuning

Configuration

To use the SDK, you need to configure it with your API key. Here's a simple setup:

import ChatGptFineTuning from 'chatgpt-fine-tuning';

const outFile = 'fine-tuning-output.jsonl'; // required
const systemMessage = 'Marv is a factual chatbot that is also sarcastic.'; // required

const gpt4Api = new ChatGptFineTuning({
    apiKey: process.env.GPT4_API_KEY || '', // required
    systemMessage,
  }, outFile)

Usage

`tuner`

Method	Parameters	Return Type	Description
`approve`	-	`Promise<void>`	Approves the current row for fine-tuning.
`reject`	-	`Promise<void>`	Rejects the current row for fine-tuning.
`fix`	`userText: string`, `assistantText: string`, `log?: boolean`	`Promise<void>`	Submits a correction for the current row, including the user and assistant text. Option to log details to file after each row.
`log`	`message: string`	`void`	Logs a message to the output file. This operation is synchronous.

Property	Type	Description
`response`	`ChatMessage`	Holds the current chat message that the tuner will operate on.

const tuner = await gpt4Api.sendMessage("What is the capital of France?");

// programmatic verification
if (tuner.response.text.includes("Paris")) {
  tuner.approve();
} else {
  // marks assistant response with a weight of 0
  tuner.reject();
  // inserts the user response and a correct assistant response with a weight of 1
  tuner.fix("You did not provide the correct answer", "Paris");
}
tuner.log("Finished run")

`ChatMessage`

Name	Type	Description
`id`	`string`	Unique identifier for the chat message.
`text`	`string`	The text content of the chat message.
`role`	`Role`	The role of the message sender (e.g., user, assistant).
`name`	`string`	The name of the sender. Optional.
`delta`	`string`	Optional string that may contain additional information or changes made to the message.
`detail`	`openai.CreateChatCompletionResponse` \| `CreateChatCompletionStreamResponse`	Optional detailed response from OpenAI or a streaming response, providing further context or metadata about the chat message.
`parentMessageId`	`string`	The ID of the parent message if this message is a reply or related to another message in the conversation. Optional.
`conversationId`	`string`	The ID of the conversation this message belongs to. Optional.