Hi all. I’m trying to estimate the cost of a fine-tuning job I’m hoping to run at some point in the future, and I was wondering if I will be billed for the tokens input/output by the model during the validation portion of supervised fine-tuning at the end of each epoch. I can’t find any mention of this in the documentation.
While my gut instinct would tell me that those tokens would be billable, when I have run fine-tuning jobs in the past the bill seemed to only include the tokens from the training file. Anything from somebody who is more familiar with the fine-tuning process than I am would be greatly appreciated!
Sorry to piggyback on this question, but when you were billed for fine-tune, how soon after the job was complete did the activity show up in your usage and billing? It’s been 2 days since my job was successfully completed and I haven’t been billed and it hasn’t shown up in my usage.
My understanding is that this is already included in the precalculated cost of training tokens for the given model, since validation loss is determined on each fine-tuning step.
In short: I wouldn’t expect to be billed. If you are billed, this isn’t documented anywhere.
Usually its near-instant for me. I know their API was having some issues earlier this week across the board, so its possible they didn’t bill for some jobs because of the delays
1 Like
Thanks for the answer! To the best of my knowledge as well after doing some more extensive research, validations tokens are not billable!