INDEX

Explanations

phrases and terms related to news articles

oai_token-act-pair · gpt-3.5-turbo

references to related topics or categories within the text

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 0-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.0.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.0.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

 humour

-0.71

 colours

-0.65

 animation

-0.65

 edges

-0.64

ises

-0.63

-0.62

 hands

-0.62

 organis

-0.62

 whistle

-0.62

ames

-0.60

POSITIVE LOGITS

RELATED

4.31

SPONSORED

1.98

Related

1.73

PHOTOS

1.65

VIDEO

1.63

READ

1.58

MORE

1.43

WATCH

1.41

ADVERTISEMENT

1.36

Advertisement

1.27

Activations Density 0.017%

phrases and terms related to news articles

references to related topics or categories within the text

No Comments

No Known Activations