INDEX

Explanations

phrases related to high-ranking officials or members of organizations, often in the context of news or issues

oai_token-act-pair · gpt-3.5-turbo

references to individuals involved in security or military contexts

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 8-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.8.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.8.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

 FANTASY

-0.92

 brunch

-0.86

 commercials

-0.79

 Nerd

-0.79

Consumer

-0.78

Apps

-0.76

flix

-0.74

Loop

-0.74

Madison

-0.74

 ballpark

-0.72

POSITIVE LOGITS

 assassinated

1.46

 Gaddafi

1.35

 assassinate

1.30

 martyr

1.28

 militant

1.19

 smugg

1.18

 Saddam

1.17

 Baghd

1.13

 Jihad

1.13

 Islamist

1.13

Activations Density 0.891%

phrases related to high-ranking officials or members of organizations, often in the context of news or issues

references to individuals involved in security or military contexts

No Comments

No Known Activations