INDEX

Explanations

phrases related to social and political actions or statements

oai_token-act-pair · gpt-3.5-turbo

repetitive phrases or expressions highlighting quantifiers and negations

oai_token-act-pair · gpt-4o-mini Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 9-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.9.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.9.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

Redditor

-0.79

 Also

-0.76

 Alternatively

-0.70

MAN

-0.67

 Additionally

-0.66

also

-0.66

 additionally

-0.65

 Nare

-0.64

Also

-0.63

rm

-0.63

POSITIVE LOGITS

whatever

0.93

etc

0.86

 EntityItem

0.75

 clot

0.68

etc

0.66

cknow

0.66

 decency

0.65

 blah

0.62

 sensit

0.62

dq

0.61

Activations Density 0.260%

phrases related to social and political actions or statements

repetitive phrases or expressions highlighting quantifiers and negations

No Comments

No Known Activations