INDEX

Explanations

words related to historical conflicts or groups associated with fighting such as "Crusaders."

oai_token-act-pair · gpt-3.5-turbo

references to "crusaders" and related terms in the context of conflict

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

 fusion

-0.70

 partner

-0.60

 cowork

-0.59

 reciproc

-0.58

 Conf

-0.57

 tele

-0.57

 Patty

-0.56

 consultation

-0.56

PH

-0.56

 Lung

-0.56

POSITIVE LOGITS

aders

4.66

ader

2.50

ading

2.41

aded

1.99

ades

1.95

asion

1.53

asions

1.43

ade

1.33

ADE

1.29

adem

1.27

Activations Density 0.013%

words related to historical conflicts or groups associated with fighting such as "Crusaders."

references to "crusaders" and related terms in the context of conflict

No Comments

No Known Activations