INDEX

Explanations

using "but" for contrast

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

******č\n

-0.11

ÂĢÂĢ

-0.10

Âģ@

-0.09

 -*-č\n

-0.09

.Formatter

-0.09

FromClass

-0.09

 jenom

-0.09

à¸·

-0.08

¨ë¶Ģ

-0.08

ķãĤĵ

-0.08

POSITIVE LOGITS

 actually

0.11

Why

0.10

 before

0.09

 wait

0.09

 instead

0.08

why

0.08

Why

0.08

 originally

0.08

 Actually

0.08

 THEN

0.08

Activations Density 1.410%