INDEX

Explanations

Transformer architecture

The neuron spotlights high-level topic or feature keywords—especially nouns used as section headings or labels (e.g. “home,” “run,” “race,” “history,” “tab,” “portability,” “pricing,” “strategy,” “station,” etc.).

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

 without

1.01

 tanpa

0.81

 zonder

0.80

 until

0.79

 без

0.77

without

0.76

 WITHOUT

0.74

 sesuai

0.74

 zgodnie

0.74

려면

0.73

POSITIVE LOGITS

which

1.38

 which

1.30

 ซึ่ง

1.29

ซึ่ง

1.24

 които

1.23

 която

1.16

 которое

1.11

 WHICH

1.10

 lequel

1.09

 które

1.07

Activations Density 0.465%