I was reading the pricing page, and it looks to me like fine-tuning a model is supposed to be free for up to 1M tokens of training data per day (at least until the end of the month). https://openai.com/api/pricing/
But when I tried to do it with about 250K tokens of training data, set for 3 epochs, it rejected me becuase it says it costs $20 to fine tune (and I currently have less than that in my account).
Are they lying on the pricing page? Any other ideas?
Which model were you trying to fine-tune?
It sounds like, even if you are not being billed, the existing rate limit and blocking based on other conditions is still in operation - seeing if you can fund the job.
You can run 1 epoch. That would reduce the cost to 1/3.
Then fine-tune again using the generated ft:model:name as the input model to base the training on. That should have similar result to the multiple passes through a training file as epoch>1 performs.
Or just feed the machine more money…