ChatGPT Hard-Fork

[mikesoylu]

Hey there love this plugin 🔥

I've spun off a hard-fork here that is based on ChatGPT and removes some functionality that you have here (diverges quite a bit from this main)

Changes introduced:

• Streaming completions with cancelling
• ChatGPT-only calls for completions & edits
• Removed some settings and loading state indicators that are no longer needed

Let me know if you'd like me to open a PR to merge those changes in 🙌

[aduros]

Hi, sounds promising! I'll play around with your fork this weekend.

How does the streaming work? It actually updates the buffer in realtime?

[mikesoylu]

Cool 🙌 Streaming uses the --no-buffer option of curl together with the api’s stream: true option and inserts every new token into the buffer as they’re available.

[aduros]

I had a go with it over the weekend, awesome work!! The streaming really makes it feel more responsive!

I like the idea of using the Chat API, since it's faster and cheaper, but the results are worse than the regular completions/edits APIs. It seems like Chat has issues where it ignores the prompt and reverts to being a virtual assistant. I think they intend to fix that in a future model though, so let's keep an eye on it... At the pace that things are going this could be an option in a month or two.

[mikesoylu]

Sounds good, I'll close this for the time being then 👍

[aduros]

I've added streaming support based on your implementation. Thanks for that!

I ended up using virtual text to display the in-progress stream rather than updating the buffer with every token, which was driving my linter plugins a bit wild 😁