INDEX

Explanations

attributes or metadata labels

The neuron strongly activates on numeric literals and their associated attribute names in structured data formats (e.g. XML/JSON), i.e. it finds numbers.

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

 will

-1.59

 your

-1.50

 from

-1.46

↵↵

-1.43

 makes

-1.41

for

-1.40

an

-1.37

 should

-1.36

 create

-1.35

at

-1.34

POSITIVE LOGITS

がありますが

1.52

了几

1.52

Recept

1.48

 ativos

1.47

 促销

1.45

Lá

1.43

MÉ

1.42

了一眼

1.42

 heridos

1.41

<bos>

1.41

Activations Density 0.010%