Is there a "moderation" endpoint for fine-tuning data?

The fine-tuning endpoint will reject certain types of content, especially for fine-tuning, that are typically not denied in other scenarios.

Is there a way to preemptively get if the fine-tuning API will reject an image or a text?

For normal Chat Completion requests, we have omni-moderation-latest. But this endpoint does not return if an image/a text is inelligible for fine-tuning.

1 Like

No there is not a moderations that aligns with what is performed on fine-tune.

It is really non-ideal to have the job proceed at your expense, after the file has been damaged, as well as receiving no report on which examples were not used.

The same goes with vision, o1 models, etc. You will be blocked, potentially striked against, but there is no prescreen mechanism.