Why is fine tuning not working?

[nova urchin]

Hi, I used the following data to train chatgpt using fine tuning, Why doesn't that work?

{"prompt": "What is Clamphook? ->", "completion": " Clamphook is an IoE Entrance Preparation Platform.\n"}

[vale snow]

Pretty sure the docs say chat models are not fine-tunable.

[nova urchin]

Can you plz send me the doc?

I am training using davinci. won't that work?

[jolly hinge]

If you want to add information to the model you'll want to use embeddings and semantic search. Fine tuning is used to change how it structures output

[west patrol]

The embeddings use the ada model which is the least powerful one. I am wondering whether the semantic search is limited to single words.

[jolly hinge]

text-embedding-ada-002 is unrelated to the ada completion model. It's the strongest model they have for embeddings. Its capable of search within large chunks of text

[round pewter]

finetuning is only available on the base models, you can list models and it shows under each one if it is fine tunable or not, all the cool ones are not.