eastriverlee / chatgpt

A clean, tiny, flexible ChatGPT asynchronous swift library that can be tailored to your need.

ChatGPT

OVERVIEW

A clean, tiny, flexible ChatGPT asynchronous library that can be tailored to your need.

USAGE

Choose one method:

• Add https://github.com/eastriverlee/ChatGPT.git using SPM using Xcode.
• Open Package.swift and add .package(url: "https://github.com/eastriverlee/ChatGPT.git", branch: "master") in dependencies.
• Copy the only source file to your project.

Base url, path, custom headers can be configured so that you can send request to not only OpenAI, but to your proxy server as well.

import ChatGPT

var openAIChatGPT = ChatGPT()
let proxiedChatGPT = ChatGPT(baseURL: "https://your.proxy.server.com", path: "v1/chatGPT")
var userAPIKey: String = "USER API KEY" { didSet { openAIChatGPT.apiKey = userAPIKey } }
var chatGPT: ChatGPT { openAIChatGPT.apiKey == nil ? proxiedChatGPT : openAIChatGPT }

BASIC

There are four types of result you can get:

1. String
2. Result
3. StringStream
4. ResultStream

Using AsyncThrowingStream's subtype StringStream, and ResultStream, you can (and probably should) get stream response without waiting for the server to come up with the whole response(which is not fast enough for users to feel comfortable waiting).

let systemPrompt = Message(role: .system, content: "You are ChatGPT, when user says something, you respond back as instructed.")

func printStringResponse(from whatYouSaid: String) async throws {
    print("You:", whatYouSaid)
    let answer = try await chatGPT.getStringResponse(from: [systemPrompt, .init(content: whatYouSaid)])
    print("ChatGPT:", answer)
}

func printResponse(from whatYouSaid: String) async throws {
    print("You:", whatYouSaid)
    let answer = try await chatGPT.getResponse(from: [systemPrompt, .init(content: whatYouSaid)])
    print("ChatGPT:", answer.choices[0].message.content)
    print("...used tokens:", answer.usage.totalTokens)
}

func printStreamResponse(from whatYouSaid: String) async throws {
    print("You:", whatYouSaid)
    let answer = try await chatGPT.getStreamResponse(from: [systemPrompt, .init(content: whatYouSaid)])
    print("ChatGPT:", terminator: " ")
    for try await answerDelta in answer {
        if let content = answerDelta.choices[0].delta.content {
            print(content, terminator: "")
            fflush(stdout)
        }
    }
}

ADVANCED

You can set model or its hyper parameters using Option struct.

func printStringStreamResponse(from whatYouSaid: String, using model: Model) async throws {
    print("You:", whatYouSaid)
    let answer = try await chatGPT.getStringStreamResponse(from: [systemPrompt, .init(content: whatYouSaid)], with: Option(model: model))
    print("ChatGPT:", terminator: " ")
    for try await answerDelta in answer {
        print(answerDelta, terminator: "")
        fflush(stdout)
    }
}

---

TEST

try await printStreamResponse(from: "Give me meaning of life in less than a hundred letters.", using: .gpt_4)

You: Give me meaning of life in less than a hundred letters.
ChatGPT: Life's meaning: seeking purpose, growth, love, and happiness while positively impacting others.

If my understanding is correct, OpenAI API does not provide information on how many tokens are used for streaming result. If you need such functionality, get help from another swift library GPT3-Tokenizer