INDEX

Explanations

measure progress and 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

 Polo

-0.10

 Tone

-0.09

rana

-0.09

ught

-0.09

arket

-0.09

FETCH

-0.09

 actionable

-0.09

TestMethod

-0.09

¶Į

-0.08

 undert

-0.08

POSITIVE LOGITS

 progress

0.22

 effectiveness

0.22

 performance

0.17

 success

0.15

progress

0.15

 Progress

0.15

 Performance

0.14

 against

0.14

performance

0.13

-progress

0.13

Activations Density 0.086%