INDEX

Explanations

negations or terms expressing denial

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

do not followed by verb

np_acts-logits-general · gemini-2.5-flash-lite

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GEMMA-2-9B-IT @ 20-gemmascope-res-131k

Configuration

google/gemma-scope-9b-it-res/layer_20/width_131k/average_l0_81

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Features

131,072

Data Type

float32

Hook Name

blocks.20.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

 somethin

-0.65

 nothin

-0.64

 outta

-0.64

 stuff

-0.63

 Aussie

-0.61

んじゃない

-0.56

-0.55

Aussie

-0.54

 nggak

-0.54

 gente

-0.53

POSITIVE LOGITS

mektedir

0.64

 sahiptir

0.58

maktadır

0.56

 volontà

0.56

 QtGui

0.55

 endast

0.54

 ElementRef

0.54

melidir

0.52

AlterField

0.52

 obtain

0.52

Activations Density 0.049%

negations or terms expressing denial

do not followed by verb

No Comments

No Known Activations