INDEX

Explanations

No Explanations Found

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 1-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.1.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.1.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

brance

-0.83

*/(

-0.79

alach

-0.78

atos

-0.76

bable

-0.74

achev

-0.74

ters

-0.71

cmp

-0.70

hedral

-0.70

thouse

-0.70

POSITIVE LOGITS

è¦ļéĨĴ

0.82

OTOS

0.77

Ranked

0.70

 Treat

0.65

Cry

0.64

âĢİ

0.64

Submit

0.63

Redditor

0.62

ye

0.62

eenth

0.62

Activations Density 0.000%

No Known Activations

This feature has no known activations.