Closed bunder2015 closed 5 months ago
I'm going to close this, I think the issue is that localai's gpt-4 implementation (which seems to be hermes-2-pro-mistral) isn't "openai-gpt4 enough". If that makes sense. :shrug:
edit: can confirm, works fine on the legit gpt-4.
Version
Command-line (Python) version
Operating System
Ubuntu Linux (docker container built on a gentoo host)
What happened?
Hello, I tried writing a prompt to generate a py+qt desktop application that involved opencv and training a new NN. Pilot didn't ask for confirmation and went straight into designing architecture and manually installing requisites which goes fine...
When it builds up a development plan it throws my entire prompt out and tries writing some boilerplate nodejs/mongodb code. I can tell it to skip all the web stuff, but then it writes a boilerplate readme and exits.
I tried a few different localai models but I can't seem to get it to actually write the code I want. I'm a little lost why it can (sometimes) read my entire description back to me word-for-word, only to ignore the whole concept of the project it was tasked to writing.
Any ideas where I can look to debug? I tried removing the files in /pilot/templates/ but that didn't seem to be enough to get it to stop writing webapps all the time.
Thanks in advance
log.txt