INDEX

Explanations

No Explanations Found

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GEMMA-2-2B @ 18-gemmascope-mlp-65k

Configuration

google/gemma-scope-2b-pt-mlp/layer_18/width_65k/average_l0_88

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Features

65,536

Data Type

float32

Hook Name

blocks.18.hook_mlp_out

Hook Layer

Architecture

jumprelu

Context Size

1,024

Dataset

monology/pile-uncopyrighted

Activation Function

relu

Embeds

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 jspb

-0.63

setVerticalGroup

-0.61

RepeatedField

-0.59

fromnode

-0.59

saraba

-0.55

 Transfermarkt

-0.53

AnimationsModule

-0.51

fromCharCode

-0.50

almaz

-0.49

 Hons

-0.49

POSITIVE LOGITS

 oprot

0.59

 تعدى

0.57

UnsafeEnabled

0.54

 дописавши

0.54

BeginContext

0.49

দ্

0.48

 فريبيس

0.48

adă

0.47

icius

0.46

typeparam

0.46

Activations Density 0.000%

No Known Activations

This feature has no known activations.