The prompt in my post was just an example it probably would work.(the prompts I actually used included proprietary/private code/data, so I cannot share them here).
Codellama is likely a lot less powerful than GPT.
Codellama is likely a lot less powerful than GPT.
iirc, there is some analysis that the alignment training nerfs model capabilities, so we are likely unintentionally making the models less capable because... FUD?
> Codellama is likely a lot less powerful than GPT.
Certainly true, but is it capable enough for your task? You'd have to try and find out. There is more analysis showing that smaller models trained for a specific task outperform large, generalist model like GPT-3.5/4