INDEX

Explanations

considered to be

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

 overwhelmingly

-0.11

perfect

-0.10

 perfect

-0.10

Perfect

-0.09

à¤°à¤¤

-0.09

 PERF

-0.09

 perfectly

-0.08

 expense

-0.08

 fines

-0.08

 Perfect

-0.08

POSITIVE LOGITS

 among

0.15

 particularly

0.13

 high

0.12

Examples

0.12

 relative

0.11

among

0.11

 candidates

0.11

 exceptions

0.11

 relatively

0.11

 reputation

0.11

Activations Density 0.160%