INDEX

Explanations

common words or names, particularly "Rut" and "jug", and some celebrity references, such as "Bruce Springsteen"

oai_token-act-pair · gpt-3.5-turbo

mentions of containers and specific names associated with them

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 5-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.5.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.5.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

 mosqu

-0.76

 spaced

-0.71

 Compton

-0.69

bom

-0.67

raid

-0.64

 flared

-0.63

 pitched

-0.63

metry

-0.62

 resistance

-0.62

 filament

-0.60

POSITIVE LOGITS

neys

1.18

glers

0.92

ebook

0.91

lain

0.82

amins

0.81

zx

0.80

geoning

0.80

secret

0.78

cle

0.75

stals

0.75

Activations Density 0.045%

common words or names, particularly "Rut" and "jug", and some celebrity references, such as "Bruce Springsteen"

mentions of containers and specific names associated with them

No Comments

No Known Activations