INDEX

Explanations

phrases containing instances of authority or control, such as instances of dictating or governing

oai_token-act-pair · gpt-3.5-turbo

terms associated with authority and control

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

trace

-0.79

 claim

-0.66

 contention

-0.62

 sight

-0.60

report

-0.60

Sou

-0.59

 prominence

-0.58

 Trouble

-0.58

 smear

-0.57

 report

-0.57

POSITIVE LOGITS

 dictated

2.58

 governed

2.28

 doomed

1.34

imov

1.22

 commanded

1.14

 indexed

1.07

ihu

1.00

 mediated

0.99

chron

0.96

 guided

0.95

Activations Density 0.047%

phrases containing instances of authority or control, such as instances of dictating or governing

terms associated with authority and control

No Comments

No Known Activations