INDEX

Explanations

phrases related to architectural features and descriptions

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

New Auto-Interp

Configuration

google/gemma-scope-9b-pt-res/layer_20/width_131k/average_l0_114

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

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

LookAnd

-0.85

__(/*!

-0.70

Diwedd

-0.65

 disambiguazione

-0.61

ItemBackground

-0.60

 Савезне

-0.59

 Reſ

-0.59

ukone

-0.57

 ſte

-0.57

wiſe

-0.57

POSITIVE LOGITS

另一个

0.35

另一個

0.34

 natomiast

0.30

 obcí

0.30

 quedado

0.29

 tijden

0.28

 interesan

0.27

 obce

0.27

 humild

0.27

 neuem

0.27

Activations Density 0.002%

phrases related to architectural features and descriptions

No Comments

No Known Activations