Do i need to improve my fine tuning?

API
1

I honestly think i suck at fine tuning

in my attempt to save money and use gpt 3.5 i thought id give fine tuning a try

I only included 10 examples in my dataset and validation

but they are full length conversations that include called functions and more

so why does my model suck so bad? lol…i thought it would tweak it to have my talking style and better understanding of when to call my functions

but instead its just…weird

is it just the size of the dataset or is there something im missing?

2

Does the prompt you are using work well with the regular gpt-3.5, i.e. the non fine-tuned model?

Generally, I would say that 10 examples is often too little to see a meaningful improvement and you should ideally start with a data set that is closer to 30 examples. That said, what’s also critical is to ensure that the prompt you are using for fine-tuning has proven to be generally effective on the regular model.