INDEX

Explanations

agent name and task

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

testdata

-0.09

 collaps

-0.08

 synd

-0.08

unn

-0.08

 Consort

-0.08

 Vice

-0.08

avis

-0.08

.createServer

-0.08

cov

-0.08

ce

-0.08

POSITIVE LOGITS

 agent

0.34

 agents

0.32

 Agents

0.27

 Agent

0.27

agent

0.26

Agent

0.25

agents

0.24

-agent

0.23

Agents

0.23

AGENT

0.21

Activations Density 0.267%