Using the OpenAI Tuning API to create a custom trained model produces rejections resulting from OpenAI usage policies with no reasonable way to track and resolve the problem. There is also apparently no appeal process to get your training data approved.
This makes the model Tuning API useless as far as I am concerned. I could never recommend it to clients who want to implement business solutions with the API.
I tried to create a harmless personal therapy Assistant by training a model from July 2024 but it will not pass the usage restrictions.
I put the same instruction prompt directly into ChatGPT and used the same prompt data, one by one, from my Tuning data set and ChatGPT responded wonderfully. But the same thing gets blocked if I try to tune (train) a model with it. Apparently the moderation API is of no help in solving the problem and only demonstrates that the usage policy doesn’t work well.
So I am left with no solution.
Anybody come up with a good answer and just their suggested “work arounds” that don’t work?