INDEX

Explanations

phrases related to news articles and their headings

oai_token-act-pair · gpt-3.5-turbo

instances of the word "RELATED" in various contexts

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 3-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.3.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.3.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

stood

-0.82

atur

-0.76

ouls

-0.75

adra

-0.73

apers

-0.73

oard

-0.71

onics

-0.70

ea

-0.69

olding

-0.69

onic

-0.67

POSITIVE LOGITS

RELATED

1.02

 IMAGES

1.00

 VIDEOS

0.98

 INFORMATION

0.90

 WATCHED

0.88

 ALSO

0.88

htaking

0.87

 APPLIC

0.86

 STOR

0.86

 STORY

0.85

Activations Density 0.006%

phrases related to news articles and their headings

instances of the word "RELATED" in various contexts

No Comments

No Known Activations