INDEX

Explanations

No Explanations Found

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 3-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.3.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.3.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

oria

-0.70

ocard

-0.69

cas

-0.68

arian

-0.67

create

-0.66

Redditor

-0.65

asty

-0.64

itor

-0.63

idon

-0.62

ca

-0.61

POSITIVE LOGITS

ntil

0.85

 mathemat

0.80

 pestic

0.79

-+-+-+-+

0.75

 expended

0.72

anooga

0.70

ntax

0.69

 maize

0.66

 challeng

0.66

 interconnected

0.65

Activations Density 0.000%

No Known Activations

This feature has no known activations.