INDEX

Explanations

references to circular objects or concepts

oai_token-act-pair · gpt-3.5-turbo

directional terms describing shapes and layouts

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

ompl

-0.80

lov

-0.78

Lib

-0.75

Vet

-0.74

 vets

-0.73

 abuser

-0.72

vet

-0.70

Nation

-0.69

tain

-0.69

afort

-0.69

POSITIVE LOGITS

 circular

2.82

 horizontal

2.60

 horizontally

2.30

 rectangular

2.01

izontal

1.97

 diagonal

1.91

 vertically

1.85

 horizont

1.78

 vertical

1.75

 triangular

1.71

Activations Density 0.032%

references to circular objects or concepts

directional terms describing shapes and layouts

No Comments

No Known Activations