INDEX

Explanations

phrases expressing preference or opposition

oai_token-act-pair · gpt-3.5-turbo

negations or phrases emphasizing the word "not."

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

eur

-0.80

velt

-0.72

kamp

-0.69

ction

-0.68

itor

-0.66

ixel

-0.65

ç·

-0.63

 Mehran

-0.63

riber

-0.62

lance

-0.62

POSITIVE LOGITS

 necessarily

1.40

icably

1.20

epad

1.09

icable

0.98

etheless

0.97

withstanding

0.94

 bothering

0.92

orious

0.88

 remotely

0.85

eworthy

0.81

Activations Density 0.091%

phrases expressing preference or opposition

negations or phrases emphasizing the word "not."

No Comments

No Known Activations