INDEX

Explanations

references to innocence or innocence being challenged

oai_token-act-pair · gpt-3.5-turbo

references to the concept of innocence, particularly in relation to individuals in precarious situations

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 0-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.0.hook_resid_pre

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

24,576

Data Type

torch.float32

Hook Point

blocks.0.hook_resid_pre

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Hook Point Layer

Activation Function

relu

Embeds

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

chart

-0.74

 Remastered

-0.70

 Scale

-0.68

 belt

-0.65

kamp

-0.64

 Migration

-0.64

 resh

-0.63

vision

-0.63

DT

-0.63

Shift

-0.63

POSITIVE LOGITS

 innocent

3.73

 innocence

2.39

 innoc

2.27

 Innocent

2.16

 unsuspecting

1.65

 innocuous

1.61

 harmless

1.60

 unwitting

1.48

 unarmed

1.47

 guilty

1.40

Activations Density 0.029%

references to innocence or innocence being challenged

references to the concept of innocence, particularly in relation to individuals in precarious situations

No Comments

No Known Activations