INDEX

Explanations

representation

The neuron activates on occurrences of the word “representation” (and its sub-parts) wherever it appears.

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

骓

-1.63

 همکاری

-1.45

 поможет

-1.45

enumerate

-1.43

 graciously

-1.38

垸

-1.38

 herr

-1.38

们的

-1.37

۸

-1.36

holdet

-1.34

POSITIVE LOGITS

 kasus

1.52

шает

1.49

</h3>

1.47

];

1.39

 Often

1.35

 delito

1.34

 juiste

1.28

ocats

1.28

ческих

1.26

荷兰

1.25

Activations Density 0.073%