INDEX

Explanations

measurement recording

The neuron fires on “code‐looking” tokens—punctuation and symbol tokens (e.g. dots, dollar signs) as well as short technical acronyms and numerals typical of programming snippets.

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

gă

-0.72

лек

-0.70

 tabla

-0.69

 chaleur

-0.69

 lachen

-0.65

VER

-0.65

 espacio

-0.64

 เว

-0.64

ContextCompat

-0.63

Uploader

-0.63

POSITIVE LOGITS

átil

0.76

별

0.69

surate

0.69

 maladies

0.67

aggregate

0.66

 infractions

0.66

面白

0.65

 Teatro

0.65

 cren

0.65

MaxSize

0.65

Activations Density 0.077%