INDEX

Explanations

keywords related to editorial content or opinions

oai_token-act-pair · gpt-3.5-turbo

references to editorial content in texts

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

 complexes

-0.81

etooth

-0.73

cles

-0.70

meg

-0.67

 drained

-0.67

heat

-0.64

»Ĵ

-0.63

0000000000000000

-0.62

 Hercules

-0.61

 taller

-0.60

POSITIVE LOGITS

 editorial

3.78

 Editorial

3.00

 editors

2.03

 Editors

1.78

 editor

1.72

editor

1.70

 journalistic

1.63

 Editor

1.43

 publishing

1.28

 journalism

1.28

Activations Density 0.015%

keywords related to editorial content or opinions

references to editorial content in texts

No Comments

No Known Activations