INDEX

Explanations

words associated with news articles or headlines

oai_token-act-pair · gpt-3.5-turbo

mentions of Washington, D.C

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 7-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.7.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.7.hook_resid_pre

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Hook Point Layer

Activation Function

relu

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

Reference

-0.70

gered

-0.68

ongo

-0.67

angers

-0.67

ads

-0.66

sole

-0.66

otropic

-0.65

Redditor

-0.64

products

-0.62

gger

-0.62

POSITIVE LOGITS

ASHINGTON

1.06

MAN

1.06

WASHINGTON

1.05

 CITY

1.03

 IMAGES

1.01

 GOODMAN

1.00

HER

1.00

 STATE

0.97

MENT

0.96

NESS

0.94

Activations Density 0.022%

words associated with news articles or headlines

mentions of Washington, D.C

No Comments

No Known Activations