hi all,
I am a reserch student and trying to train/fine tune babbage-002 with 800 datapoints. I got following warning after failed fine tune status:
"This file failed moderation safety checks. The OpenAI Moderation API identifies fine tuning examples that violate our content policies. " My question is :
if my data has original user posts (prompt)containing such language which is helpful in identifying mental health root cause(completions) classification job. Then how to pass this stage?
ANY advise please. I am really stuck at this stage.
there is no advise unless you contact support, for permission because it got flagged by OpenAI Moderation safety checks
1 Like
What you can do is pull out the examples and send them to the moderations endpoint yourself, and record the scores and flagging.
You may discover with analysis both the type of inputs or desired responses that are triggering, and which others might contribute to a total “reject” score.
1 Like
thnak you , will do so… let see if i can get arround as most of my content has such converstaion, which are people real life experience contributing to their mental health.