INDEX

Explanations

account related tokens

The neuron fires on technical jargon tokens—especially uppercase acronyms and code identifiers (e.g. EFI, UEFI, BC, module or account names) used in protocols, APIs, or file formats.

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

gazione

-1.65

alted

-1.53

escence

-1.47

ンダ

-1.47

ământ

-1.41

いました

-1.38

شود

-1.34

textsc

-1.30

ză

-1.29

 estén

-1.27

POSITIVE LOGITS

1.73

1.52

初め

1.43

 douane

1.39

 than

1.38

𝐃

1.38

：《

1.34

↳

1.29

↵↵

1.27

Activations Density 0.007%