INDEX

Explanations

news-related phrases

oai_token-act-pair · gpt-3.5-turbo

phrases and words that introduce or emphasize parts of an argument or statement

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 5-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.5.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.5.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

 helicop

-0.59

 Vaugh

-0.59

 destro

-0.55

 challeng

-0.53

 blah

-0.51

farious

-0.51

 advoc

-0.51

enegger

-0.51

 behav

-0.50

 âĢ¦"

-0.50

POSITIVE LOGITS

resa

0.88

xiety

0.76

 largeDownload

0.65

BSD

0.63

odore

0.60

Berry

0.59

asionally

0.57

tymology

0.56

ories

0.56

ropolitan

0.53

Activations Density 0.438%

news-related phrases

phrases and words that introduce or emphasize parts of an argument or statement

No Comments

No Known Activations