INDEX

Explanations

code diffs and code context

This neuron activates on numeric metadata in unified diffs—line numbers, hunk ranges, indexes, timestamps, and other numeric tokens in patch headers.

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

of

-1.45

 строительство

-1.34

”،

-1.33

የ

-1.30

 hatta

-1.30

 rojas

-1.27

更多的

-1.25

Ꮉ

-1.23

足够的

-1.23

 вони

-1.23

POSITIVE LOGITS

the

1.88

 when

1.66

 first

1.53

when

1.49

one

1.48

thirds

1.46

⁺

1.44

 Einzelnen

1.43

 quando

1.37

>−

1.36

Activations Density 0.045%