INDEX

Explanations

code diffs

The neuron fires on numeric tokens—particularly digit sequences used as version numbers, line numbers, timestamps, or other numeric identifiers in code diffs.

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

its

-1.41

of

-1.27

 just

-1.24

 each

-1.24

by

-1.18

as

-1.15

or

-1.13

 their

-1.12

 over

-1.11

 after

-1.08

POSITIVE LOGITS

んですよ

1.30

should

1.19

 nieuwe

1.16

に入れる

1.11

 nuovi

1.09

Should

1.06

 größe

1.06

からお

1.06

 grueso

1.06

様に

1.05

Activations Density 0.008%