INDEX

Explanations

Prefixes forming words

This neuron fires on the initial title-style tokens (the first few letters of words) that appear as section or heading markers at the start of blocks.

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

tonsoft

-1.50

quelize

-1.43

 schwarze

-1.41

ίος

-1.38

řízen

-1.38

ș

-1.37

 weißem

-1.36

 וכן

-1.32

—.

-1.30

érale

-1.30

POSITIVE LOGITS

2.23

to

2.00

ing

1.57

’

1.42

 aéri

1.38

onTap

1.34

ないけど

1.30

vč

1.30

 find

1.27

～、

1.27

Activations Density 0.101%