INDEX

Explanations

phrases related to instructional or advisory content

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GEMMA-2-2B @ 6-gemmascope-mlp-16k

Configuration

google/gemma-scope-2b-pt-mlp/layer_6/width_16k/average_l0_133

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Features

16,384

Data Type

float32

Hook Name

blocks.6.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

extAlignment

-0.65

 insuffisamment

-0.59

AndEndTag

-0.55

 human

-0.54

cticut

-0.50

ViewImports

-0.50

EndInit

-0.48

she

-0.47

xffffffff

-0.47

BindView

-0.47

POSITIVE LOGITS

 Eurasia

0.63

 GenerationType

0.60

 morire

0.59

 exportación

0.57

 avoient

0.57

 africains

0.56

 démocr

0.56

 Herodotus

0.55

webElementGuid

0.55

 Pollut

0.54

Activations Density 0.996%

phrases related to instructional or advisory content

No Comments

No Known Activations