3.0.0

Note
Fun Fact: Parts of the documentation for this update have been written by GPT4! I have proof-read them to make sure everything is fine but GPT4 just did it perfectly.

Added

• Added GPT4 models (also including stable and large context versions) to ChatGPT. If you have access, enjoy!
• Added GPT, which is a wrapper around the completion API for GPT3. Also supports streaming the answers.
• Added OpenAI, which is a wrapper around two general OpenAI endpoints that list available models.
• Added more robust and convenient error handling. All errors are now exposed through GPTSwiftError.
  • OpenAI error messages are now also passed through, which is useful to debug invalid requests (like out-of-bounds values).
• Added option to configure underlying URLSession in the initializers of ChatGPT, GPT and OpenAI.
• Added option to set the default model to use for instances of ChatGPT, GPT and OpenAI. You can set that in the initializer and then, if needed, override it in the individual methods.
• Added convenience methods to initialize requests and messages, for example ChatRequest.gpt3(_:) or ChatMessage.system(_:).
• Added curl(for:pretty:formatOutput:) methods to ChatGPT and GPT that turns requests into a usable curl prompt that you can paste into a terminal to manually call the OpenAI endpoints.
  • OpenAI also has these convenience methods: curlForAvailableModels(pretty:formatOutput:) and curlForModel(withId:pretty:formatOutput:)

Changed

• Renamed ChatGPTSwift to ChatGPT.
• ChatGPT.ask(_:) and ChatGPT.ask(messages:) have been made easier to use. They now return a String as the answer, instead of a complex object.
  • The intention behind this is to make basic usage of the framework as easy as possible while still allowing full control through ChatGPT.ask(request:)
• ChatGPT, GPT and OpenAI now each have their own product to reduce namespace pollution. Simply import whichever you are using.