INDEX

Explanations

sequences and letters

The neuron strongly activates on tokens that come from the tail end of the character set (e.g. letters o–z and digits 8–9), i.e. it flags high‐indexed/vocabulary‐end characters.

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

様な

-1.27

ごとの

-1.13

感じです

-1.05

際に

-1.05

 arī

-1.01

感じが

-1.01

為に

-1.01

optarg

-0.98

際の

-0.97

>);

-0.96

POSITIVE LOGITS

<bos>

1.31

かで

1.30

がありますが

1.09

百貨

1.09

もしか

1.06

があり

1.06

社会主义

1.04

つまり

1.02

<0xA3>

1.00

也就是说

1.00

Activations Density 0.018%