INDEX

Explanations

the word "re" followed by a high level of activation

oai_token-act-pair · gpt-3.5-turbo

the word "we" in various contexts, indicating a collective perspective or shared experiences

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

 lapt

-0.80

BuyableInstoreAndOnline

-0.69

owl

-0.63

 Seah

-0.62

 Pebble

-0.62

 freezes

-0.61

 simulator

-0.61

 layer

-0.60

 deterrent

-0.59

 Bernstein

-0.59

POSITIVE LOGITS

ngth

1.05

aper

1.03

apers

0.99

selves

0.85

becca

0.83

versible

0.83

claimer

0.82

acters

0.80

psons

0.79

ggie

0.78

Activations Density 0.037%

the word "re" followed by a high level of activation

the word "we" in various contexts, indicating a collective perspective or shared experiences

No Comments

No Known Activations