INDEX

Explanations

specific values or words

The neuron strongly activates on subword pieces that form the endings of longer, multi-syllable or technical words (i.e. it detects common word suffixes or final segments).

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

only

-1.38

 only

-1.37

褡

-1.30

 subsequently

-1.30

ஂ

-1.27

 Jazeera

-1.24

 merely

-1.21

 LAURA

-1.21

 stets

-1.19

Whenever

-1.19

POSITIVE LOGITS

мне

1.48

 inicia

1.36

</em>

1.34

 пиво

1.32

 hasty

1.30

Dě

1.28

совет

1.26

обы

1.21

здрав

1.20

доступ

1.20

Activations Density 0.109%