INDEX

Explanations

ensure things become

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

affer

-0.09

 still

-0.09

 contrast

-0.09

 Fallen

-0.09

 jadx

-0.09

 consect

-0.09

ardy

-0.09

 æħ

-0.08

-have

-0.08

 appropriate

-0.08

POSITIVE LOGITS

 becomes

0.16

 become

0.16

 serving

0.13

 serves

0.13

 serve

0.13

æĪĲä¸º

0.12

 truly

0.12

serve

0.11

 slou

0.11

erves

0.11

Activations Density 0.065%