Comment by bigmadshoe

So it is tuned specifically to classify content for safeguarding? I'm not familiar with this particular model, but it most likely has a specific classifier head that is tuned for the safeguard task. This is completely different to zero-shot classification.