What does the histogram in feature info page indicate?

[ptyin]

I cannot find explanation for that figure. Is that a Feature Density Histogram?

[ptyin]

Got it. The x axis indicates activation value, and the y is for token counts, right?

[zdaiot]

@ptyin I think so too. Do you know what "Prediction prior to ablated token" means? Why do most of the words here show as "Prediction prior to ablated token", and only a small part show as KL divergence?

[WuTheFWasThat]

@ptyin yes that's correct!

@zdaiot "Prediction prior to ablated token" is because we're showing effects when ablating the boxed token. All predictions of tokens up to the ablated token are unaffected