layer 28 activation on different model sizes - does the uncertainty signal show up in 7b and 13b too

seeing layer 28 fire high before hallucinations on Llama 3.3 70b, wondering if this pattern holds on smaller models or if it's specific to 70b scale. if the uncertainty signal is a real emergent property we should see it on 7b and 13b too, maybe at different layer positions (scaled by depth). if it only shows up at 70b that's interesting for interpretability research has anyone tested activation patterns on smaller Llama 3 models before hallucination?