INDEX

Explanations

phrases indicating attachment or connection between objects or entities

oai_token-act-pair · gpt-3.5-turbo

instances of the word "attached" and variations related to attachment

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

izz

-0.70

nuts

-0.64

iasco

-0.63

fund

-0.61

bia

-0.60

apeake

-0.59

geries

-0.59

 homeland

-0.59

 Marathon

-0.58

iably

-0.58

POSITIVE LOGITS

 attached

3.80

 attach

2.04

 attaching

1.94

 attachments

1.94

 attachment

1.92

 attaches

1.86

Attach

1.65

 embedded

1.52

 detached

1.38

 enclosed

1.37

Activations Density 0.005%

phrases indicating attachment or connection between objects or entities

instances of the word "attached" and variations related to attachment

No Comments

No Known Activations