INDEX

Explanations

effect or effectiveness

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

ed

-0.14

adients

-0.09

aret

-0.09

erc

-0.09

GEST

-0.09

 Clint

-0.09

 Charm

-0.09

imoto

-0.09

 Wheeler

-0.08

POSITIVE LOGITS

uate

0.26

ual

0.26

ively

0.25

ually

0.24

iveness

0.24

uated

0.24

ives

0.22

uating

0.22

uation

0.21

uates

0.19

Activations Density 0.021%