INDEX

Explanations

terms related to preventing harm or risk in various contexts

oai_token-act-pair · gpt-4o-mini Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GEMMA-2-9B @ 38-gemmascope-res-16k

Configuration

google/gemma-scope-9b-pt-res/layer_38/width_16k/average_l0_128

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Features

16,384

Data Type

float32

Hook Name

blocks.38.hook_resid_post

Hook Layer

Architecture

jumprelu

Context Size

1,024

Dataset

monology/pile-uncopyrighted

Activation Function

relu

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

abstractmethod

-0.47

bedingt

-0.46

cleaning

-0.45

 Everywhere

-0.44

 Reisedaten

-0.43

 cleaning

-0.42

 examining

-0.41

parenting

-0.41

Everywhere

-0.41

everywhere

-0.41

POSITIVE LOGITS

 future

0.85

 harm

0.80

 potential

0.80

 possible

0.80

 futuras

0.75

 posibles

0.75

 unnecessary

0.73

 futuros

0.71

future

0.70

 accidental

0.70

Activations Density 0.613%

terms related to preventing harm or risk in various contexts

No Comments

No Known Activations