INDEX

Explanations

The neuron strongly fires on the very first subword or word piece that follows punctuation (hyphens, apostrophes, question marks, etc.).

at least one

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

♭

-1.24

 stadion

-1.23

>+</

-1.20

rages

-1.15

'#

-1.15

 lande

-1.14

UCIÓN

-1.13

 brune

-1.13

 '+':

-1.13

 tillbaka

-1.12

POSITIVE LOGITS

を行います

1.45

に行って

1.16

its

1.14

Unless

1.08

はずです

1.07

を開始

1.06

one

1.05

we

1.05

entreprises

1.05

 wasn

1.04

Activations Density 0.037%