INDEX

Explanations

news headlines or titles containing the word "WATCH"

oai_token-act-pair · gpt-3.5-turbo

instances of video content or specific watch-related cues

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

 Shape

-0.72

ound

-0.64

abase

-0.62

Sin

-0.62

 Inher

-0.60

aries

-0.57

 inherit

-0.57

 delinqu

-0.56

acle

-0.56

erk

-0.55

POSITIVE LOGITS

WATCH

3.94

VIDEO

2.14

LIST

1.42

BRE

1.36

READ

1.24

 WATCH

1.24

PHOTOS

1.22

TIME

1.19

RELATED

1.16

CHECK

1.13

Activations Density 0.028%

news headlines or titles containing the word "WATCH"

instances of video content or specific watch-related cues

No Comments

No Known Activations