INDEX

Explanations

encoder-decoder structure

The neuron flags content‐bearing nouns—especially proper names and specialized topic terms.

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

同士

0.66

ప్పటికీ

0.59

与之

0.59

 অথচ

0.58

eri

0.57

 मुझसे

0.55

ত্ব

0.55

钕

0.54

abatic

0.53

ورية

0.52

POSITIVE LOGITS

 ratio

1.34

 combination

1.34

 combo

1.31

 combinatie

1.27

combination

1.25

 relationship

1.20

ratio

1.18

combo

1.16

 Kombination

1.15

relationship

1.14

Activations Density 0.328%