Code generation is not up to mark

[SBTopZZZ-LG]

Hi. I don't know if this is the expected behaviour of the model, but the code it generates is not up to mark.

I was testing and goofing around with it for quite some time now, and it's very decent. Later however, I decided to ask it to write some codes for me and, it seems like the model only generates codes in Python 3, regardless of the specified syntax through the prompt.

Is this an expected behaviour? Thank you.

[imwide]

Try rerunning the programm. It might get stuck on a behaviour it showed in previous messages.

[mio9]

Actually if you look at what the models are trained on at the botton of the website here https://gpt4all.io , you will notice python (the orange and red dots) is involved a lot, and its presence is so large it's categorized by itself, and so model thinks python is the way to go whenever programming is mentioned. It definitely is not a desirable behavior, but it's also expected from how the models are made.

So, yes, it is expected, you can try forcing your language by ending your prompt with ```java to signal this java code block has to be there, and hopefully the model will start to make more sense on what you're asking for.

[mio9]

Quick update, I tested with Java using GPT4All-L13b-snoozy , seems like it's working just as well after assisting the model for a bit

EDIT: Same approach works for other languages, but the model prefer using pre-implemented functions rather than actually chewing out algorithms

[niansa]

Stale, please open a new, updated issue if this is still relevant.