Switched back to the completion API using the gpt-3.5-turbo-instruct model.
It worked almost out of the box, but there was a strange issue where,
when triggered on a line with indentation, it keeps adding a newline and the same indentation
(so almost all completions start with an empty line).
I wrote a postprocessing function in LeapInterfaces.ts to fix that before getting to the user.
Please test it out (it is deployed on the web) and merge if you're happy with it :)
Switched back to the completion API using the
gpt-3.5-turbo-instructmodel.It worked almost out of the box, but there was a strange issue where, when triggered on a line with indentation, it keeps adding a newline and the same indentation (so almost all completions start with an empty line). I wrote a postprocessing function in
LeapInterfaces.tsto fix that before getting to the user.Please test it out (it is deployed on the web) and merge if you're happy with it :)