INDEX

Explanations

technical commit messages

This neuron detects numeric tokens (digits, version numbers, and other standalone numbers).

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

 передний

-1.38

or

-1.34

so

-1.27

 with

-1.23

for

-1.21

 decorativos

-1.16

 very

-1.16

TWO

-1.13

 when

-1.12

 where

-1.12

POSITIVE LOGITS

 executar

1.51

paravant

1.48

 geprüft

1.46

 ativar

1.38

Farbe

1.35

Mitar

1.33

 limpar

1.32

mostly

1.30

znych

1.29

INCLUDING

1.29

Activations Density 0.022%