INDEX

Explanations

balance, path, tensor, updates

The neuron never activates on any tokens here, suggesting it’s effectively unused (i.e. a “dead” neuron that doesn’t detect any pattern in these documents).

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

 handlerInput

0.92

 parâmetros

0.91

 físicos

0.89

 específicos

0.88

 marít

0.88

 espectadores

0.87

 núcleos

0.87

énieur

0.86

 этот

0.85

 слегка

0.84

POSITIVE LOGITS

こと

0.73

‘

0.67

0.66

 import

0.66

0.64

pan

0.62

『

0.61

"/

0.61

近年来

0.60

Activations Density 0.001%