INDEX

Explanations

uncommon or obscure

New Auto-Interp

Top Features by Cosine Similarity

Configuration

Prompts (Dashboard)

10,000 prompts, 128 tokens each

Dataset (Dashboard)

lmsys/lmsys-chat-1m

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 desn

-0.09

noop

-0.09

 unrelated

-0.09

ahl

-0.09

recur

-0.08

 Willi

-0.08

gum

-0.08

ohan

-0.08

.:.:

-0.08

ovnÄĽ

-0.08

POSITIVE LOGITS

 lesser

0.29

 obscure

0.29

 niche

0.28

 rare

0.26

obs

0.26

 uncommon

0.23

 Lesser

0.22

 less

0.20

 unusual

0.19

 smaller

0.19

Activations Density 0.258%