INDEX

Explanations

names related to a specific individual, possibly related to a news or legal context

oai_token-act-pair · gpt-3.5-turbo

proper nouns and names, particularly referring to a specific individual

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

 Bore

-0.81

 Witt

-0.80

Ney

-0.76

Lov

-0.76

 Davidson

-0.75

 Ober

-0.74

Station

-0.74

minster

-0.73

 Square

-0.73

 Franch

-0.73

POSITIVE LOGITS

 peel

1.64

Mal

1.21

 mantle

1.19

isman

1.18

 Vick

1.17

Xin

1.15

 minion

1.13

 Ming

1.06

 Snapdragon

1.04

 Cait

1.02

Activations Density 0.068%

names related to a specific individual, possibly related to a news or legal context

proper nouns and names, particularly referring to a specific individual

No Comments

No Known Activations