INDEX

Explanations

proper nouns, potentially related to news articles or other text comprising of a mixture of letters and numbers

oai_token-act-pair · gpt-3.5-turbo

specific acronyms or abbreviations often related to organizations or government entities

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 1-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.1.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.1.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

hol

-0.91

fe

-0.82

iors

-0.81

itant

-0.74

aign

-0.74

omore

-0.74

ite

-0.74

gard

-0.73

hor

-0.73

auga

-0.73

POSITIVE LOGITS

IRO

1.66

 IMAGES

1.30

IRED

1.29

ION

1.28

ITAL

1.27

ORN

1.25

ELY

1.20

ECT

1.19

ATOR

1.19

IS

1.18

Activations Density 0.026%

proper nouns, potentially related to news articles or other text comprising of a mixture of letters and numbers

specific acronyms or abbreviations often related to organizations or government entities

No Comments

No Known Activations