INDEX

Explanations

finalization and conclusions

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

Zen

-0.10

old

-0.10

ed

-0.10

 grap

-0.10

 Wahl

-0.09

.cloudflare

-0.09

ideon

-0.09

 finalized

-0.09

edge

-0.09

POSITIVE LOGITS

izing

0.26

ised

0.26

ization

0.23

ity

0.23

izes

0.21

mente

0.21

izer

0.19

izers

0.19

ising

0.19

ized

0.19

Activations Density 0.018%