INDEX

Explanations

the word 'par' with varying activation strengths, potentially indicating a focus on specific text related to certain topics or entities

oai_token-act-pair · gpt-3.5-turbo

references to paragraphs or sections within a document

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 3-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.3.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.3.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

ħĭ

-0.94

¥µ

-0.86

doms

-0.75

æ©Ł

-0.68

ĨĴ

-0.68

 customary

-0.67

urst

-0.67

OME

-0.65

å¸

-0.65

houses

-0.63

POSITIVE LOGITS

allel

1.20

liament

1.03

anoia

0.93

imony

0.89

vati

0.87

cel

0.80

ity

0.80

ippi

0.79

amount

0.78

ion

0.78

Activations Density 0.006%

the word 'par' with varying activation strengths, potentially indicating a focus on specific text related to certain topics or entities

references to paragraphs or sections within a document

No Comments

No Known Activations