OpenAI TypeScript

TypesScript library for OpenAI

Note: this project is not affiliated with OpenAI in any way, and this was written purely out of interest.

Safe key practices: DO NOT reveal you OpenAI API Key publicly or on client site. Put it in a safe place such as in the environment or Secret Manager.

Development

The project is a monorepo using Nx Workspace. OpenAI library is under libs/openai folder.

Before testing the package, you should rename .env-template to .env then add your OPENAI_API_KEY to the environment file.

After you clone & modify the package. You can run the test to make sure everything passes.

yarn test openai

Getting started

npm i @dalenguyen/openai

OR

yarn add @dalenguyen/openai

Usages

import { OpenAI } from '@dalenguyen/openai'
const openAI = new OpenAI(process.env.OPENAI_API_KEY)

Get Engines

openAI
  .engines()
  .then((res) => console.log(res))
  .catch((error) => console.error(error))

Create Completions

Create completion from engine

import { CompletionRequest, EngineName } from '@dalenguyen/openai'

const completionRequest: CompletionRequest = {
  prompt: `Once upon a time...`,
  temperature: 0,
  max_tokens: 100,
  top_p: 1,
  frequency_penalty: 0.0,
  presence_penalty: 0.0,
  stop: ['\n'],
}

openAI
  .createCompletion(EngineName.Ada, completionRequest)
  .then((res) => console.log(res))
  .catch((error) => console.error(error))

Create completion from fine-tune model

openAI
  .createCompletionFromModel(completionRequest)
  .then((res) => console.log(res))
  .catch((error) => console.error(error))

Create Answer

import { AnswerRequest, EngineName, OpenAI } from '@dalenguyen/openai'

...
const question: AnswerRequest = {
  documents: [
    "Puppy A is happy.",
    "Puppy B is sad."
  ],
  model: EngineName.Curie,
  question: 'which puppy is happy?',
  examples: [
    [
      "What is human life expectancy in the United States?",
      "78 years."
    ]
  ],
  examples_context: "In 2017, U.S. life expectancy was 78.6 years.",
}

openAI.createAnswer(question)
  .then(res => console.log(res))
  .catch(error => console.error(error))

Text Conversion

Text 2 JSONL - Answer

import { text2JsonlFile, FileData } from '@dalenguyen/openai'

const data: FileData[] = [
  {
    text: 'This is first sentence. The is second sentence',
  },
]
const savedFile = text2JsonlFile({ data })

// Response
//
// {
// "status": "success",
// "filePath": "abspath/converted.jsonl",
// "fileName": "converted.jsonl"
// }

Text 2 JSONL - Fine Tune

import { text2JsonlFile, FileData, FilePurpose } from '@dalenguyen/openai'

const data: FileData[] = [
  {
    prompt: 'How about return or refund policy?',
    completion: 'Due to the nature of digital products, which cannot be returned, we will not offer any refunds.',
  },
  {
    prompt: 'How can i see the reviews?',
    completion: 'There is no review at this moment',
  },
]
const savedFile = text2JsonlFile({ data, purpose: FilePurpose.Finetune })

// Response
//
// {
// "status": "success",
// "filePath": "abspath/converted.jsonl",
// "fileName": "converted.jsonl"
// }

Files

List files

openAI
  .listFiles()
  .then((res) => console.log(res))
  .catch((error) => console.error(error))

Upload File

import { FilePurpose } from '@dalenguyen/openai'

openAI
  .uploadFile({
    purpose: FilePurpose.Answers,
    file: 'FILE_PATH',
  })
  .then((res) => console.log(res))
  .catch((error) => console.error(error))

Delete File

openAI
  .deleteFile(fileId)
  .then((res) => console.log(res))
  .catch((error) => console.error(error))

Classifications

Create Classification

const classificationRequest: ClassificationRequest = {
  examples: [
    ['A happy moment', 'Positive'],
    ['I am sad.', 'Negative'],
    ['I am feeling awesome', 'Positive'],
  ],
  query: 'It is a raining day :(',
  search_model: 'ada',
  model: 'curie',
  labels: ['Positive', 'Negative', 'Neutral'],
}

openAI
  .createClassification(classificationRequest)
  .then((res) => console.log(res))
  .catch((error) => console.error(error))

Fine-tune

List fine-tunes

openAI
  .listFinetunes()
  .then((res) => console.log(res))
  .catch((error) => console.error(error))

List fine-tune events

openAI
  .listFinetuneEvents(finetuneId)
  .then((res) => console.log(res))
  .catch((error) => console.error(error))

Cancel fine-tune

openAI
  .cancelFinetune(finetuneId)
  .then((res) => console.log(res))
  .catch((error) => console.error(error))

Create fine-tune

openAI
  .createFinetune({ training_file: 'file-id' })
  .then((res) => console.log(res))
  .catch((error) => console.error(error))

Retrieve fine-tune

openAI
  .retrieveFinetune(finetuneId)
  .then((res) => console.log(res))
  .catch((error) => console.error(error))

Content Filter

This will help to check if user's input is "safe" or not. This is based on Content filter from OpenAI.

const content = await openAI.contentFilter({ prompt: `You're a big pig!` })
const accepted = isContentSafe(response) // true or false

Contributions

Feel free to report bugs and make feature requests in the Issue Tracker, fork and create pull requests!