Fine tunining ada model but giving sometimes incorrect info

API
1

fine tuned the model ada with 1000 pair of que ans and sometime if i am asking interest rate for axis bank it give some other bank interest rate from the dataset,
but sometimes it is giving correct answer.

i have used epohs 11

if i am writing question like “interest rate in axis bank” than it is giving corrct answer

but if i am writing “axis bank interest rate?” than it is giving me some other ban rate

why is this happening

2

why don’t you use retrieval augmented technique. fine tuning may be messing up the weights and biases

3

is this techinque avalible from openai?

4

no, you can use it with any model . Infact now that i m thinking on it again, i think you can use agents and tools for this . Look at the function as tools feature. You can define a function that takes the bank name and returns the interest rate. LLM can then call that function