INDEX

Explanations

once the

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

-0.08

erule

-0.08

aris

-0.08

 already

-0.08

ares

-0.08

Nic

-0.08

 ØªØ±ÛĮÙĨ

-0.08

ahi

-0.08

-valu

-0.08

 Cors

-0.08

POSITIVE LOGITS

 initial

0.18

 dust

0.16

 desired

0.15

ses

0.15

 above

0.14

initial

0.13

 basic

0.13

 basics

0.12

 initially

0.12

desired

0.12

Activations Density 0.016%