INDEX

Explanations

references to helpfulness or supportive information

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

 pouvoit

-0.73

 étoit

-0.72

 auroit

-0.71

 avoient

-0.68

 Houſe

-0.64

 feroit

-0.63

 Inscrivez

-0.63

Spoljašnje

-0.61

Хьажоргаш

-0.61

matchCondition

-0.61

POSITIVE LOGITS

 Helpful

0.79

 helpful

0.76

Helpful

0.71

helpful

0.69

orable

0.54

ductory

0.47

 pertinent

0.47

pilar

0.47

responsive

0.46

tuous

0.46

Activations Density 0.011%

references to helpfulness or supportive information

No Comments

No Known Activations