INDEX

Explanations

prefixes of words or names

The neuron fires on capitalized subword tokens that begin proper names or acronyms.

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

 that

-1.55

 protože

-1.50

had

-1.45

can

-1.37

That

-1.32

)$,

-1.30

 можли

-1.30

 might

-1.29

そろ

-1.27

 щодо

-1.23

POSITIVE LOGITS

 både

1.63

both

1.30

Væ

1.27

 netts

1.24

wut

1.23

 pria

1.23

yup

1.23

 both

1.22

;"

1.21

 verschillende

1.21

Activations Density 0.788%