Training private version of model

Hi everyone,

a newbie who is trying to understand what this brilliant service is all about. There are some questions that I can’t seem to pinpoint any answer and would really appreciate community’s kind feedback.

I can see that we can train the base model with our private data using fine-tunings and embeddings (still trying to understand which one is the best option) but if we do train the model, do we get a private version of model like model ID or some sort of other information that can only be used by certain users and not accessed publicly?

The use case I am trying to implement is that we have large legal documents (can be 100 or so pages in some cases) and we want to build an interface for our users to select a particular document and then ask questions in the context of that document only.

All models that are fine-tuned are private to the account that requested the fine-tune.

Only the account that requested fine-tuned model has access to it and controls access to it, and yes you get a whole new fine_tuned_model like ada:ft-your-org:custom-model-name-2022-02-15-04-21-04

I would definitely recommend reading through docs on fine-tuning and embeddings before you proceed.

On a first look, it appears like embeddings could do the job for this case, but depends on the details.

Do you have developers on your team?


Thanks a lot for your kind response. Going through the links that you shared.

I am a developer myself (web developer so not very good on data and AI bits) but doing a bit of initial feasibility to see if openAI fits our use case.


If you search the forums, you’ll find several threads regarding fine-tuning and uses.

Good luck on your quest!