INDEX

Explanations

names or places after photo:

This neuron detects image‐related metadata tokens (e.g. “Photo,” “Image,” photographer names, credit lines, and associated numbers).

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

 本店

-1.48

 alfombra

-1.47

 artikelen

-1.43

 tratando

-1.42

 asegurarse

-1.41

 préférence

-1.41

 lámparas

-1.40

が入

-1.39

bandeja

-1.39

 matériaux

-1.38

POSITIVE LOGITS

}%

1.60

柞

1.53

kowej

1.52

ieke

1.50

 ممکن

1.49

UnifiedTopology

1.46

ТОВ

1.45

Ꮞ

1.45

ﺶ

1.44

 poden

1.41

Activations Density 0.015%