INDEX

Explanations

narrowing down

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

 flattened

-0.11

 weakening

-0.10

 lowering

-0.10

 LOWER

-0.10

 Minh

-0.10

 Lower

-0.09

 lowers

-0.09

 Depths

-0.09

Nh

-0.09

 flatt

-0.09

POSITIVE LOGITS

 narrow

0.40

 Narrow

0.37

nar

0.30

 narrowed

0.29

 ì¢

0.27

Nar

0.24

nar

0.24

wh

0.23

 narrower

0.23

ç¼©

0.23

Activations Density 0.184%