INDEX

Explanations

compound words with specific prefixes

The neuron strongly activates on standalone numeric or hex‐style tokens (i.e. literal numbers, addresses, or other digit/letter sequences used as constants).

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

 вважа

-0.93

ESTRO

-0.77

jame

-0.75

GRU

-0.72

CHOOL

-0.72

 kaynağından

-0.71

spects

-0.71

ocino

-0.71

 vierte

-0.71

ginas

-0.70

POSITIVE LOGITS

 перед

0.85

tests

0.76

Datenschutz

0.75

test

0.75

ิจ

0.75

prepend

0.74

Werbung

0.73

Aggregator

0.72

Lah

0.71

anzia

0.71

Activations Density 0.001%