INDEX

Explanations

case settlements reported examine

This neuron fires on numeric tokens—e.g. counts, percentages, years, measurements—i.e. it spots numbers/statistical data in the text.

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

 with

-1.16

Descriere

-1.14

 without

-1.13

 because

-1.12

for

-1.07

Câ

-1.06

劑

-1.04

嗰

-1.04

or

-1.03

you

-1.02

POSITIVE LOGITS

 saveurs

1.40

וּ

1.27

 untersucht

1.23

 fuß

1.23

ansı

1.22

 content

1.19

significantly

1.17

 vélos

1.14

 across

1.12

 бампер

1.11

Activations Density 0.079%