INDEX

Explanations

classifying levels or types

This neuron fires on short, all-caps or code-style tokens (acronyms, model numbers, resolution tags, initials, etc.)—in other words, technical abbreviations and similar uppercase “coded” terms.

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

 plays

0.69

 güçlü

0.61

蛋白质

0.60

ভিতে

0.60

 avulla

0.58

പരമായ

0.57

ভালো

0.57

 tenfold

0.56

 magnetism

0.55

 rationing

0.54

POSITIVE LOGITS

،

1.59

 және

1.49

 그리고

1.47

1.45

၊

1.43

-,

1.40

 และ

1.40

‌,

1.40

 そして

1.38

 এবং

1.36

Activations Density 0.275%