DreamAI
I created a simple Javascript library called "DreamAI". Its sole purpose is to interact easily with the OpenAI API LLM (Large Language Model) using asynchronous calls. This code is free to any to use, but please keep the header intact.
/*
* DreamAI is a lightweight Javascript class designed to interact with the OpenAI API for completions and images.
* To register for an OpenAI account, please go to: https://platform.openai.com/
* Once you have your API key, you can use these methods to generate completions and images
* using the language model of your choice.
* DreamAI is (c) Matthew Martin, 2023.
* https://www.dreaminhex.com
*/
class DreamAI {
/* Available (most common) language models for OpenAI */
/* See: https://platform.openai.com/docs/models/gpt-3 */
/* https://platform.openai.com/docs/models/gpt-4 */
static get GPT4() { return "gpt-4"; }
static get GPT35() { return "gpt-3.5-turbo"; }
static get DaVinci003() { return "text-davinci-003"; }
static get DaVinci002() { return "text-davinci-002"; }
static get DaVinci001() { return "text-davinci-001"; }
static get Curie001() { return "text-curie-001"; }
static get Babbage001() { return "text-babbage-001"; }
static get Ada001() { return "text-ada-001"; }
/* Construct the DreamAI class using your OpenAPI key, and provide the model you wish to use. Additional parameters
* can be used to tune the model to respond in a specific way.
*/
constructor(openAIKey, model, options) {
this.openAIKey = openAIKey;
this.model = model;
if (this.model[0] !== 'g') {
this.temperature = options.temperature;
this.top_p = options.top_p;
this.frequency_penalty = options.frequency_penalty;
this.presence_penalty = options.presence_penalty;
}
}
/* Provide a prompt, and training data, and the system will return a message. */
async Complete(prompt, trainingData, callback) {
let messages = [];
messages[0] =
{
role: "system",
content: trainingData.join(' '),
};
messages[1] =
{
role: "assistant",
content: prompt,
};
let request =
{
messages: messages,
model: this.model,
max_tokens: this.model[0] === 'g' ? 4096 : 512
};
// Only use the options settings for non-GPT models.
if (this.model[0] !== 'g') {
request.temperature = this.temperature === null ? 0.7 : this.temperature;
request.top_p = this.top_p === null ? 1 : this.top_p;
request.frequency_penalty = this.frequency_penalty === null ? 0 : this.frequency_penalty;
request.presence_penalty = this.presence_penalty === null ? 0 : this.presence_penalty;
}
let data = await this.#postAsync("https://api.openai.com/v1/chat/completions", request);
if (data) {
var responseMessage = data.choices[0].message.content;
callback(responseMessage);
}
else {
console.log("DreamAI: No data was returned. Check your API key.");
}
}
/* Provide a prompt and image size to return a single URL of the generated image. */
async Image(prompt, width, height, callback) {
await this.Images(prompt, width, height, 1, callback);
}
/* Provide a prompt, image size, and count of images (no more than 10) to return a list of image URLs. */
async Images(prompt, width, height, count, callback) {
if (count > 10) {
console.log("DreamAI: No more than 10 images are allowed.");
}
else {
let request =
{
prompt: this.#formatPrompt(prompt),
n: count,
size: width + "x" + height
};
let data = await this.#postAsync("https://api.openai.com/v1/images/generations", request, callback);
if (data) {
if (count === 1) {
var imageUrl = data.data[0].url;
callback(imageUrl);
}
else {
var urls = [];
data.data.map((i) => urls.push(i.url));
callback(urls);
}
}
else {
console.log("DreamAI: No data was returned. Check your API key.");
}
}
}
#formatPrompt(prompt) {
if (prompt.length <= 400) {
return prompt;
}
else {
return prompt.substring(0, 399);
}
}
async #postAsync(url, json) {
return await fetch(url, {
method: 'POST',
headers: {
'Accept': 'application/json',
'Content-Type': 'application/json',
'Authorization': 'Bearer ' + this.openAIKey
},
body: JSON.stringify(json)
})
.then((response) => response.json())
.then(data => {
return data;
})
.catch(error => {
console.error(error);
});
}
}